FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_umtx.c
1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2015, 2016 The FreeBSD Foundation
5 * Copyright (c) 2004, David Xu <davidxu@freebsd.org>
6 * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
7 * All rights reserved.
8 *
9 * Portions of this software were developed by Konstantin Belousov
10 * under sponsorship from the FreeBSD Foundation.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice unmodified, this list of conditions, and the following
17 * disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_umtx_profiling.h"
38
39 #include <sys/param.h>
40 #include <sys/kernel.h>
41 #include <sys/fcntl.h>
42 #include <sys/file.h>
43 #include <sys/filedesc.h>
44 #include <sys/limits.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mman.h>
48 #include <sys/mutex.h>
49 #include <sys/priv.h>
50 #include <sys/proc.h>
51 #include <sys/resource.h>
52 #include <sys/resourcevar.h>
53 #include <sys/rwlock.h>
54 #include <sys/sbuf.h>
55 #include <sys/sched.h>
56 #include <sys/smp.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/systm.h>
60 #include <sys/sysproto.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/taskqueue.h>
63 #include <sys/time.h>
64 #include <sys/eventhandler.h>
65 #include <sys/umtx.h>
66
67 #include <security/mac/mac_framework.h>
68
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_object.h>
74
75 #include <machine/atomic.h>
76 #include <machine/cpu.h>
77
78 #include <compat/freebsd32/freebsd32.h>
79 #ifdef COMPAT_FREEBSD32
80 #include <compat/freebsd32/freebsd32_proto.h>
81 #endif
82
83 #define _UMUTEX_TRY 1
84 #define _UMUTEX_WAIT 2
85
86 #ifdef UMTX_PROFILING
87 #define UPROF_PERC_BIGGER(w, f, sw, sf) \
88 (((w) > (sw)) || ((w) == (sw) && (f) > (sf)))
89 #endif
90
91 /* Priority inheritance mutex info. */
92 struct umtx_pi {
93 /* Owner thread */
94 struct thread *pi_owner;
95
96 /* Reference count */
97 int pi_refcount;
98
99 /* List entry to link umtx holding by thread */
100 TAILQ_ENTRY(umtx_pi) pi_link;
101
102 /* List entry in hash */
103 TAILQ_ENTRY(umtx_pi) pi_hashlink;
104
105 /* List for waiters */
106 TAILQ_HEAD(,umtx_q) pi_blocked;
107
108 /* Identify a userland lock object */
109 struct umtx_key pi_key;
110 };
111
112 /* A userland synchronous object user. */
113 struct umtx_q {
114 /* Linked list for the hash. */
115 TAILQ_ENTRY(umtx_q) uq_link;
116
117 /* Umtx key. */
118 struct umtx_key uq_key;
119
120 /* Umtx flags. */
121 int uq_flags;
122 #define UQF_UMTXQ 0x0001
123
124 /* The thread waits on. */
125 struct thread *uq_thread;
126
127 /*
128 * Blocked on PI mutex. read can use chain lock
129 * or umtx_lock, write must have both chain lock and
130 * umtx_lock being hold.
131 */
132 struct umtx_pi *uq_pi_blocked;
133
134 /* On blocked list */
135 TAILQ_ENTRY(umtx_q) uq_lockq;
136
137 /* Thread contending with us */
138 TAILQ_HEAD(,umtx_pi) uq_pi_contested;
139
140 /* Inherited priority from PP mutex */
141 u_char uq_inherited_pri;
142
143 /* Spare queue ready to be reused */
144 struct umtxq_queue *uq_spare_queue;
145
146 /* The queue we on */
147 struct umtxq_queue *uq_cur_queue;
148 };
149
150 TAILQ_HEAD(umtxq_head, umtx_q);
151
152 /* Per-key wait-queue */
153 struct umtxq_queue {
154 struct umtxq_head head;
155 struct umtx_key key;
156 LIST_ENTRY(umtxq_queue) link;
157 int length;
158 };
159
160 LIST_HEAD(umtxq_list, umtxq_queue);
161
162 /* Userland lock object's wait-queue chain */
163 struct umtxq_chain {
164 /* Lock for this chain. */
165 struct mtx uc_lock;
166
167 /* List of sleep queues. */
168 struct umtxq_list uc_queue[2];
169 #define UMTX_SHARED_QUEUE 0
170 #define UMTX_EXCLUSIVE_QUEUE 1
171
172 LIST_HEAD(, umtxq_queue) uc_spare_queue;
173
174 /* Busy flag */
175 char uc_busy;
176
177 /* Chain lock waiters */
178 int uc_waiters;
179
180 /* All PI in the list */
181 TAILQ_HEAD(,umtx_pi) uc_pi_list;
182
183 #ifdef UMTX_PROFILING
184 u_int length;
185 u_int max_length;
186 #endif
187 };
188
189 #define UMTXQ_LOCKED_ASSERT(uc) mtx_assert(&(uc)->uc_lock, MA_OWNED)
190
191 /*
192 * Don't propagate time-sharing priority, there is a security reason,
193 * a user can simply introduce PI-mutex, let thread A lock the mutex,
194 * and let another thread B block on the mutex, because B is
195 * sleeping, its priority will be boosted, this causes A's priority to
196 * be boosted via priority propagating too and will never be lowered even
197 * if it is using 100%CPU, this is unfair to other processes.
198 */
199
200 #define UPRI(td) (((td)->td_user_pri >= PRI_MIN_TIMESHARE &&\
201 (td)->td_user_pri <= PRI_MAX_TIMESHARE) ?\
202 PRI_MAX_TIMESHARE : (td)->td_user_pri)
203
204 #define GOLDEN_RATIO_PRIME 2654404609U
205 #ifndef UMTX_CHAINS
206 #define UMTX_CHAINS 512
207 #endif
208 #define UMTX_SHIFTS (__WORD_BIT - 9)
209
210 #define GET_SHARE(flags) \
211 (((flags) & USYNC_PROCESS_SHARED) == 0 ? THREAD_SHARE : PROCESS_SHARE)
212
213 #define BUSY_SPINS 200
214
215 struct abs_timeout {
216 int clockid;
217 bool is_abs_real; /* TIMER_ABSTIME && CLOCK_REALTIME* */
218 struct timespec cur;
219 struct timespec end;
220 };
221
222 struct umtx_copyops {
223 int (*copyin_timeout)(const void *uaddr, struct timespec *tsp);
224 int (*copyin_umtx_time)(const void *uaddr, size_t size,
225 struct _umtx_time *tp);
226 int (*copyin_robust_lists)(const void *uaddr, size_t size,
227 struct umtx_robust_lists_params *rbp);
228 int (*copyout_timeout)(void *uaddr, size_t size,
229 struct timespec *tsp);
230 const size_t timespec_sz;
231 const size_t umtx_time_sz;
232 const bool compat32;
233 };
234
235 _Static_assert(sizeof(struct umutex) == sizeof(struct umutex32), "umutex32");
236 _Static_assert(__offsetof(struct umutex, m_spare[0]) ==
237 __offsetof(struct umutex32, m_spare[0]), "m_spare32");
238
239 int umtx_shm_vnobj_persistent = 0;
240 SYSCTL_INT(_kern_ipc, OID_AUTO, umtx_vnode_persistent, CTLFLAG_RWTUN,
241 &umtx_shm_vnobj_persistent, 0,
242 "False forces destruction of umtx attached to file, on last close");
243 static int umtx_max_rb = 1000;
244 SYSCTL_INT(_kern_ipc, OID_AUTO, umtx_max_robust, CTLFLAG_RWTUN,
245 &umtx_max_rb, 0,
246 "Maximum number of robust mutexes allowed for each thread");
247
248 static uma_zone_t umtx_pi_zone;
249 static struct umtxq_chain umtxq_chains[2][UMTX_CHAINS];
250 static MALLOC_DEFINE(M_UMTX, "umtx", "UMTX queue memory");
251 static int umtx_pi_allocated;
252
253 static SYSCTL_NODE(_debug, OID_AUTO, umtx, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
254 "umtx debug");
255 SYSCTL_INT(_debug_umtx, OID_AUTO, umtx_pi_allocated, CTLFLAG_RD,
256 &umtx_pi_allocated, 0, "Allocated umtx_pi");
257 static int umtx_verbose_rb = 1;
258 SYSCTL_INT(_debug_umtx, OID_AUTO, robust_faults_verbose, CTLFLAG_RWTUN,
259 &umtx_verbose_rb, 0,
260 "");
261
262 #ifdef UMTX_PROFILING
263 static long max_length;
264 SYSCTL_LONG(_debug_umtx, OID_AUTO, max_length, CTLFLAG_RD, &max_length, 0, "max_length");
265 static SYSCTL_NODE(_debug_umtx, OID_AUTO, chains, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
266 "umtx chain stats");
267 #endif
268
269 static void abs_timeout_update(struct abs_timeout *timo);
270
271 static void umtx_shm_init(void);
272 static void umtxq_sysinit(void *);
273 static void umtxq_hash(struct umtx_key *key);
274 static struct umtxq_chain *umtxq_getchain(struct umtx_key *key);
275 static void umtxq_unlock(struct umtx_key *key);
276 static void umtxq_busy(struct umtx_key *key);
277 static void umtxq_unbusy(struct umtx_key *key);
278 static void umtxq_insert_queue(struct umtx_q *uq, int q);
279 static void umtxq_remove_queue(struct umtx_q *uq, int q);
280 static int umtxq_sleep(struct umtx_q *uq, const char *wmesg, struct abs_timeout *);
281 static int umtxq_count(struct umtx_key *key);
282 static struct umtx_pi *umtx_pi_alloc(int);
283 static void umtx_pi_free(struct umtx_pi *pi);
284 static int do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags,
285 bool rb);
286 static void umtx_thread_cleanup(struct thread *td);
287 SYSINIT(umtx, SI_SUB_EVENTHANDLER+1, SI_ORDER_MIDDLE, umtxq_sysinit, NULL);
288
289 #define umtxq_signal(key, nwake) umtxq_signal_queue((key), (nwake), UMTX_SHARED_QUEUE)
290 #define umtxq_insert(uq) umtxq_insert_queue((uq), UMTX_SHARED_QUEUE)
291 #define umtxq_remove(uq) umtxq_remove_queue((uq), UMTX_SHARED_QUEUE)
292
293 static struct mtx umtx_lock;
294
295 #ifdef UMTX_PROFILING
296 static void
297 umtx_init_profiling(void)
298 {
299 struct sysctl_oid *chain_oid;
300 char chain_name[10];
301 int i;
302
303 for (i = 0; i < UMTX_CHAINS; ++i) {
304 snprintf(chain_name, sizeof(chain_name), "%d", i);
305 chain_oid = SYSCTL_ADD_NODE(NULL,
306 SYSCTL_STATIC_CHILDREN(_debug_umtx_chains), OID_AUTO,
307 chain_name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
308 "umtx hash stats");
309 SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
310 "max_length0", CTLFLAG_RD, &umtxq_chains[0][i].max_length, 0, NULL);
311 SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
312 "max_length1", CTLFLAG_RD, &umtxq_chains[1][i].max_length, 0, NULL);
313 }
314 }
315
316 static int
317 sysctl_debug_umtx_chains_peaks(SYSCTL_HANDLER_ARGS)
318 {
319 char buf[512];
320 struct sbuf sb;
321 struct umtxq_chain *uc;
322 u_int fract, i, j, tot, whole;
323 u_int sf0, sf1, sf2, sf3, sf4;
324 u_int si0, si1, si2, si3, si4;
325 u_int sw0, sw1, sw2, sw3, sw4;
326
327 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
328 for (i = 0; i < 2; i++) {
329 tot = 0;
330 for (j = 0; j < UMTX_CHAINS; ++j) {
331 uc = &umtxq_chains[i][j];
332 mtx_lock(&uc->uc_lock);
333 tot += uc->max_length;
334 mtx_unlock(&uc->uc_lock);
335 }
336 if (tot == 0)
337 sbuf_printf(&sb, "%u) Empty ", i);
338 else {
339 sf0 = sf1 = sf2 = sf3 = sf4 = 0;
340 si0 = si1 = si2 = si3 = si4 = 0;
341 sw0 = sw1 = sw2 = sw3 = sw4 = 0;
342 for (j = 0; j < UMTX_CHAINS; j++) {
343 uc = &umtxq_chains[i][j];
344 mtx_lock(&uc->uc_lock);
345 whole = uc->max_length * 100;
346 mtx_unlock(&uc->uc_lock);
347 fract = (whole % tot) * 100;
348 if (UPROF_PERC_BIGGER(whole, fract, sw0, sf0)) {
349 sf0 = fract;
350 si0 = j;
351 sw0 = whole;
352 } else if (UPROF_PERC_BIGGER(whole, fract, sw1,
353 sf1)) {
354 sf1 = fract;
355 si1 = j;
356 sw1 = whole;
357 } else if (UPROF_PERC_BIGGER(whole, fract, sw2,
358 sf2)) {
359 sf2 = fract;
360 si2 = j;
361 sw2 = whole;
362 } else if (UPROF_PERC_BIGGER(whole, fract, sw3,
363 sf3)) {
364 sf3 = fract;
365 si3 = j;
366 sw3 = whole;
367 } else if (UPROF_PERC_BIGGER(whole, fract, sw4,
368 sf4)) {
369 sf4 = fract;
370 si4 = j;
371 sw4 = whole;
372 }
373 }
374 sbuf_printf(&sb, "queue %u:\n", i);
375 sbuf_printf(&sb, "1st: %u.%u%% idx: %u\n", sw0 / tot,
376 sf0 / tot, si0);
377 sbuf_printf(&sb, "2nd: %u.%u%% idx: %u\n", sw1 / tot,
378 sf1 / tot, si1);
379 sbuf_printf(&sb, "3rd: %u.%u%% idx: %u\n", sw2 / tot,
380 sf2 / tot, si2);
381 sbuf_printf(&sb, "4th: %u.%u%% idx: %u\n", sw3 / tot,
382 sf3 / tot, si3);
383 sbuf_printf(&sb, "5th: %u.%u%% idx: %u\n", sw4 / tot,
384 sf4 / tot, si4);
385 }
386 }
387 sbuf_trim(&sb);
388 sbuf_finish(&sb);
389 sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
390 sbuf_delete(&sb);
391 return (0);
392 }
393
394 static int
395 sysctl_debug_umtx_chains_clear(SYSCTL_HANDLER_ARGS)
396 {
397 struct umtxq_chain *uc;
398 u_int i, j;
399 int clear, error;
400
401 clear = 0;
402 error = sysctl_handle_int(oidp, &clear, 0, req);
403 if (error != 0 || req->newptr == NULL)
404 return (error);
405
406 if (clear != 0) {
407 for (i = 0; i < 2; ++i) {
408 for (j = 0; j < UMTX_CHAINS; ++j) {
409 uc = &umtxq_chains[i][j];
410 mtx_lock(&uc->uc_lock);
411 uc->length = 0;
412 uc->max_length = 0;
413 mtx_unlock(&uc->uc_lock);
414 }
415 }
416 }
417 return (0);
418 }
419
420 SYSCTL_PROC(_debug_umtx_chains, OID_AUTO, clear,
421 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
422 sysctl_debug_umtx_chains_clear, "I",
423 "Clear umtx chains statistics");
424 SYSCTL_PROC(_debug_umtx_chains, OID_AUTO, peaks,
425 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0,
426 sysctl_debug_umtx_chains_peaks, "A",
427 "Highest peaks in chains max length");
428 #endif
429
430 static void
431 umtxq_sysinit(void *arg __unused)
432 {
433 int i, j;
434
435 umtx_pi_zone = uma_zcreate("umtx pi", sizeof(struct umtx_pi),
436 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
437 for (i = 0; i < 2; ++i) {
438 for (j = 0; j < UMTX_CHAINS; ++j) {
439 mtx_init(&umtxq_chains[i][j].uc_lock, "umtxql", NULL,
440 MTX_DEF | MTX_DUPOK);
441 LIST_INIT(&umtxq_chains[i][j].uc_queue[0]);
442 LIST_INIT(&umtxq_chains[i][j].uc_queue[1]);
443 LIST_INIT(&umtxq_chains[i][j].uc_spare_queue);
444 TAILQ_INIT(&umtxq_chains[i][j].uc_pi_list);
445 umtxq_chains[i][j].uc_busy = 0;
446 umtxq_chains[i][j].uc_waiters = 0;
447 #ifdef UMTX_PROFILING
448 umtxq_chains[i][j].length = 0;
449 umtxq_chains[i][j].max_length = 0;
450 #endif
451 }
452 }
453 #ifdef UMTX_PROFILING
454 umtx_init_profiling();
455 #endif
456 mtx_init(&umtx_lock, "umtx lock", NULL, MTX_DEF);
457 umtx_shm_init();
458 }
459
460 struct umtx_q *
461 umtxq_alloc(void)
462 {
463 struct umtx_q *uq;
464
465 uq = malloc(sizeof(struct umtx_q), M_UMTX, M_WAITOK | M_ZERO);
466 uq->uq_spare_queue = malloc(sizeof(struct umtxq_queue), M_UMTX,
467 M_WAITOK | M_ZERO);
468 TAILQ_INIT(&uq->uq_spare_queue->head);
469 TAILQ_INIT(&uq->uq_pi_contested);
470 uq->uq_inherited_pri = PRI_MAX;
471 return (uq);
472 }
473
474 void
475 umtxq_free(struct umtx_q *uq)
476 {
477
478 MPASS(uq->uq_spare_queue != NULL);
479 free(uq->uq_spare_queue, M_UMTX);
480 free(uq, M_UMTX);
481 }
482
483 static inline void
484 umtxq_hash(struct umtx_key *key)
485 {
486 unsigned n;
487
488 n = (uintptr_t)key->info.both.a + key->info.both.b;
489 key->hash = ((n * GOLDEN_RATIO_PRIME) >> UMTX_SHIFTS) % UMTX_CHAINS;
490 }
491
492 static inline struct umtxq_chain *
493 umtxq_getchain(struct umtx_key *key)
494 {
495
496 if (key->type <= TYPE_SEM)
497 return (&umtxq_chains[1][key->hash]);
498 return (&umtxq_chains[0][key->hash]);
499 }
500
501 /*
502 * Lock a chain.
503 *
504 * The code is a macro so that file/line information is taken from the caller.
505 */
506 #define umtxq_lock(key) do { \
507 struct umtx_key *_key = (key); \
508 struct umtxq_chain *_uc; \
509 \
510 _uc = umtxq_getchain(_key); \
511 mtx_lock(&_uc->uc_lock); \
512 } while (0)
513
514 /*
515 * Unlock a chain.
516 */
517 static inline void
518 umtxq_unlock(struct umtx_key *key)
519 {
520 struct umtxq_chain *uc;
521
522 uc = umtxq_getchain(key);
523 mtx_unlock(&uc->uc_lock);
524 }
525
526 /*
527 * Set chain to busy state when following operation
528 * may be blocked (kernel mutex can not be used).
529 */
530 static inline void
531 umtxq_busy(struct umtx_key *key)
532 {
533 struct umtxq_chain *uc;
534
535 uc = umtxq_getchain(key);
536 mtx_assert(&uc->uc_lock, MA_OWNED);
537 if (uc->uc_busy) {
538 #ifdef SMP
539 if (smp_cpus > 1) {
540 int count = BUSY_SPINS;
541 if (count > 0) {
542 umtxq_unlock(key);
543 while (uc->uc_busy && --count > 0)
544 cpu_spinwait();
545 umtxq_lock(key);
546 }
547 }
548 #endif
549 while (uc->uc_busy) {
550 uc->uc_waiters++;
551 msleep(uc, &uc->uc_lock, 0, "umtxqb", 0);
552 uc->uc_waiters--;
553 }
554 }
555 uc->uc_busy = 1;
556 }
557
558 /*
559 * Unbusy a chain.
560 */
561 static inline void
562 umtxq_unbusy(struct umtx_key *key)
563 {
564 struct umtxq_chain *uc;
565
566 uc = umtxq_getchain(key);
567 mtx_assert(&uc->uc_lock, MA_OWNED);
568 KASSERT(uc->uc_busy != 0, ("not busy"));
569 uc->uc_busy = 0;
570 if (uc->uc_waiters)
571 wakeup_one(uc);
572 }
573
574 static inline void
575 umtxq_unbusy_unlocked(struct umtx_key *key)
576 {
577
578 umtxq_lock(key);
579 umtxq_unbusy(key);
580 umtxq_unlock(key);
581 }
582
583 static struct umtxq_queue *
584 umtxq_queue_lookup(struct umtx_key *key, int q)
585 {
586 struct umtxq_queue *uh;
587 struct umtxq_chain *uc;
588
589 uc = umtxq_getchain(key);
590 UMTXQ_LOCKED_ASSERT(uc);
591 LIST_FOREACH(uh, &uc->uc_queue[q], link) {
592 if (umtx_key_match(&uh->key, key))
593 return (uh);
594 }
595
596 return (NULL);
597 }
598
599 static inline void
600 umtxq_insert_queue(struct umtx_q *uq, int q)
601 {
602 struct umtxq_queue *uh;
603 struct umtxq_chain *uc;
604
605 uc = umtxq_getchain(&uq->uq_key);
606 UMTXQ_LOCKED_ASSERT(uc);
607 KASSERT((uq->uq_flags & UQF_UMTXQ) == 0, ("umtx_q is already on queue"));
608 uh = umtxq_queue_lookup(&uq->uq_key, q);
609 if (uh != NULL) {
610 LIST_INSERT_HEAD(&uc->uc_spare_queue, uq->uq_spare_queue, link);
611 } else {
612 uh = uq->uq_spare_queue;
613 uh->key = uq->uq_key;
614 LIST_INSERT_HEAD(&uc->uc_queue[q], uh, link);
615 #ifdef UMTX_PROFILING
616 uc->length++;
617 if (uc->length > uc->max_length) {
618 uc->max_length = uc->length;
619 if (uc->max_length > max_length)
620 max_length = uc->max_length;
621 }
622 #endif
623 }
624 uq->uq_spare_queue = NULL;
625
626 TAILQ_INSERT_TAIL(&uh->head, uq, uq_link);
627 uh->length++;
628 uq->uq_flags |= UQF_UMTXQ;
629 uq->uq_cur_queue = uh;
630 return;
631 }
632
633 static inline void
634 umtxq_remove_queue(struct umtx_q *uq, int q)
635 {
636 struct umtxq_chain *uc;
637 struct umtxq_queue *uh;
638
639 uc = umtxq_getchain(&uq->uq_key);
640 UMTXQ_LOCKED_ASSERT(uc);
641 if (uq->uq_flags & UQF_UMTXQ) {
642 uh = uq->uq_cur_queue;
643 TAILQ_REMOVE(&uh->head, uq, uq_link);
644 uh->length--;
645 uq->uq_flags &= ~UQF_UMTXQ;
646 if (TAILQ_EMPTY(&uh->head)) {
647 KASSERT(uh->length == 0,
648 ("inconsistent umtxq_queue length"));
649 #ifdef UMTX_PROFILING
650 uc->length--;
651 #endif
652 LIST_REMOVE(uh, link);
653 } else {
654 uh = LIST_FIRST(&uc->uc_spare_queue);
655 KASSERT(uh != NULL, ("uc_spare_queue is empty"));
656 LIST_REMOVE(uh, link);
657 }
658 uq->uq_spare_queue = uh;
659 uq->uq_cur_queue = NULL;
660 }
661 }
662
663 /*
664 * Check if there are multiple waiters
665 */
666 static int
667 umtxq_count(struct umtx_key *key)
668 {
669 struct umtxq_queue *uh;
670
671 UMTXQ_LOCKED_ASSERT(umtxq_getchain(key));
672 uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE);
673 if (uh != NULL)
674 return (uh->length);
675 return (0);
676 }
677
678 /*
679 * Check if there are multiple PI waiters and returns first
680 * waiter.
681 */
682 static int
683 umtxq_count_pi(struct umtx_key *key, struct umtx_q **first)
684 {
685 struct umtxq_queue *uh;
686
687 *first = NULL;
688 UMTXQ_LOCKED_ASSERT(umtxq_getchain(key));
689 uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE);
690 if (uh != NULL) {
691 *first = TAILQ_FIRST(&uh->head);
692 return (uh->length);
693 }
694 return (0);
695 }
696
697 /*
698 * Wake up threads waiting on an userland object.
699 */
700
701 static int
702 umtxq_signal_queue(struct umtx_key *key, int n_wake, int q)
703 {
704 struct umtxq_queue *uh;
705 struct umtx_q *uq;
706 int ret;
707
708 ret = 0;
709 UMTXQ_LOCKED_ASSERT(umtxq_getchain(key));
710 uh = umtxq_queue_lookup(key, q);
711 if (uh != NULL) {
712 while ((uq = TAILQ_FIRST(&uh->head)) != NULL) {
713 umtxq_remove_queue(uq, q);
714 wakeup(uq);
715 if (++ret >= n_wake)
716 return (ret);
717 }
718 }
719 return (ret);
720 }
721
722 /*
723 * Wake up specified thread.
724 */
725 static inline void
726 umtxq_signal_thread(struct umtx_q *uq)
727 {
728
729 UMTXQ_LOCKED_ASSERT(umtxq_getchain(&uq->uq_key));
730 umtxq_remove(uq);
731 wakeup(uq);
732 }
733
734 static inline int
735 tstohz(const struct timespec *tsp)
736 {
737 struct timeval tv;
738
739 TIMESPEC_TO_TIMEVAL(&tv, tsp);
740 return tvtohz(&tv);
741 }
742
743 static void
744 abs_timeout_init(struct abs_timeout *timo, int clockid, int absolute,
745 const struct timespec *timeout)
746 {
747
748 timo->clockid = clockid;
749 if (!absolute) {
750 timo->is_abs_real = false;
751 abs_timeout_update(timo);
752 timespecadd(&timo->cur, timeout, &timo->end);
753 } else {
754 timo->end = *timeout;
755 timo->is_abs_real = clockid == CLOCK_REALTIME ||
756 clockid == CLOCK_REALTIME_FAST ||
757 clockid == CLOCK_REALTIME_PRECISE;
758 /*
759 * If is_abs_real, umtxq_sleep will read the clock
760 * after setting td_rtcgen; otherwise, read it here.
761 */
762 if (!timo->is_abs_real) {
763 abs_timeout_update(timo);
764 }
765 }
766 }
767
768 static void
769 abs_timeout_init2(struct abs_timeout *timo, const struct _umtx_time *umtxtime)
770 {
771
772 abs_timeout_init(timo, umtxtime->_clockid,
773 (umtxtime->_flags & UMTX_ABSTIME) != 0, &umtxtime->_timeout);
774 }
775
776 static inline void
777 abs_timeout_update(struct abs_timeout *timo)
778 {
779
780 kern_clock_gettime(curthread, timo->clockid, &timo->cur);
781 }
782
783 static int
784 abs_timeout_gethz(struct abs_timeout *timo)
785 {
786 struct timespec tts;
787
788 if (timespeccmp(&timo->end, &timo->cur, <=))
789 return (-1);
790 timespecsub(&timo->end, &timo->cur, &tts);
791 return (tstohz(&tts));
792 }
793
794 static uint32_t
795 umtx_unlock_val(uint32_t flags, bool rb)
796 {
797
798 if (rb)
799 return (UMUTEX_RB_OWNERDEAD);
800 else if ((flags & UMUTEX_NONCONSISTENT) != 0)
801 return (UMUTEX_RB_NOTRECOV);
802 else
803 return (UMUTEX_UNOWNED);
804
805 }
806
807 /*
808 * Put thread into sleep state, before sleeping, check if
809 * thread was removed from umtx queue.
810 */
811 static inline int
812 umtxq_sleep(struct umtx_q *uq, const char *wmesg, struct abs_timeout *abstime)
813 {
814 struct umtxq_chain *uc;
815 int error, timo;
816
817 if (abstime != NULL && abstime->is_abs_real) {
818 curthread->td_rtcgen = atomic_load_acq_int(&rtc_generation);
819 abs_timeout_update(abstime);
820 }
821
822 uc = umtxq_getchain(&uq->uq_key);
823 UMTXQ_LOCKED_ASSERT(uc);
824 for (;;) {
825 if (!(uq->uq_flags & UQF_UMTXQ)) {
826 error = 0;
827 break;
828 }
829 if (abstime != NULL) {
830 timo = abs_timeout_gethz(abstime);
831 if (timo < 0) {
832 error = ETIMEDOUT;
833 break;
834 }
835 } else
836 timo = 0;
837 error = msleep(uq, &uc->uc_lock, PCATCH | PDROP, wmesg, timo);
838 if (error == EINTR || error == ERESTART) {
839 umtxq_lock(&uq->uq_key);
840 break;
841 }
842 if (abstime != NULL) {
843 if (abstime->is_abs_real)
844 curthread->td_rtcgen =
845 atomic_load_acq_int(&rtc_generation);
846 abs_timeout_update(abstime);
847 }
848 umtxq_lock(&uq->uq_key);
849 }
850
851 curthread->td_rtcgen = 0;
852 return (error);
853 }
854
855 /*
856 * Convert userspace address into unique logical address.
857 */
858 int
859 umtx_key_get(const void *addr, int type, int share, struct umtx_key *key)
860 {
861 struct thread *td = curthread;
862 vm_map_t map;
863 vm_map_entry_t entry;
864 vm_pindex_t pindex;
865 vm_prot_t prot;
866 boolean_t wired;
867
868 key->type = type;
869 if (share == THREAD_SHARE) {
870 key->shared = 0;
871 key->info.private.vs = td->td_proc->p_vmspace;
872 key->info.private.addr = (uintptr_t)addr;
873 } else {
874 MPASS(share == PROCESS_SHARE || share == AUTO_SHARE);
875 map = &td->td_proc->p_vmspace->vm_map;
876 if (vm_map_lookup(&map, (vm_offset_t)addr, VM_PROT_WRITE,
877 &entry, &key->info.shared.object, &pindex, &prot,
878 &wired) != KERN_SUCCESS) {
879 return (EFAULT);
880 }
881
882 if ((share == PROCESS_SHARE) ||
883 (share == AUTO_SHARE &&
884 VM_INHERIT_SHARE == entry->inheritance)) {
885 key->shared = 1;
886 key->info.shared.offset = (vm_offset_t)addr -
887 entry->start + entry->offset;
888 vm_object_reference(key->info.shared.object);
889 } else {
890 key->shared = 0;
891 key->info.private.vs = td->td_proc->p_vmspace;
892 key->info.private.addr = (uintptr_t)addr;
893 }
894 vm_map_lookup_done(map, entry);
895 }
896
897 umtxq_hash(key);
898 return (0);
899 }
900
901 /*
902 * Release key.
903 */
904 void
905 umtx_key_release(struct umtx_key *key)
906 {
907 if (key->shared)
908 vm_object_deallocate(key->info.shared.object);
909 }
910
911 #ifdef COMPAT_FREEBSD10
912 /*
913 * Lock a umtx object.
914 */
915 static int
916 do_lock_umtx(struct thread *td, struct umtx *umtx, u_long id,
917 const struct timespec *timeout)
918 {
919 struct abs_timeout timo;
920 struct umtx_q *uq;
921 u_long owner;
922 u_long old;
923 int error = 0;
924
925 uq = td->td_umtxq;
926 if (timeout != NULL)
927 abs_timeout_init(&timo, CLOCK_REALTIME, 0, timeout);
928
929 /*
930 * Care must be exercised when dealing with umtx structure. It
931 * can fault on any access.
932 */
933 for (;;) {
934 /*
935 * Try the uncontested case. This should be done in userland.
936 */
937 owner = casuword(&umtx->u_owner, UMTX_UNOWNED, id);
938
939 /* The acquire succeeded. */
940 if (owner == UMTX_UNOWNED)
941 return (0);
942
943 /* The address was invalid. */
944 if (owner == -1)
945 return (EFAULT);
946
947 /* If no one owns it but it is contested try to acquire it. */
948 if (owner == UMTX_CONTESTED) {
949 owner = casuword(&umtx->u_owner,
950 UMTX_CONTESTED, id | UMTX_CONTESTED);
951
952 if (owner == UMTX_CONTESTED)
953 return (0);
954
955 /* The address was invalid. */
956 if (owner == -1)
957 return (EFAULT);
958
959 error = thread_check_susp(td, false);
960 if (error != 0)
961 break;
962
963 /* If this failed the lock has changed, restart. */
964 continue;
965 }
966
967 /*
968 * If we caught a signal, we have retried and now
969 * exit immediately.
970 */
971 if (error != 0)
972 break;
973
974 if ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK,
975 AUTO_SHARE, &uq->uq_key)) != 0)
976 return (error);
977
978 umtxq_lock(&uq->uq_key);
979 umtxq_busy(&uq->uq_key);
980 umtxq_insert(uq);
981 umtxq_unbusy(&uq->uq_key);
982 umtxq_unlock(&uq->uq_key);
983
984 /*
985 * Set the contested bit so that a release in user space
986 * knows to use the system call for unlock. If this fails
987 * either some one else has acquired the lock or it has been
988 * released.
989 */
990 old = casuword(&umtx->u_owner, owner, owner | UMTX_CONTESTED);
991
992 /* The address was invalid. */
993 if (old == -1) {
994 umtxq_lock(&uq->uq_key);
995 umtxq_remove(uq);
996 umtxq_unlock(&uq->uq_key);
997 umtx_key_release(&uq->uq_key);
998 return (EFAULT);
999 }
1000
1001 /*
1002 * We set the contested bit, sleep. Otherwise the lock changed
1003 * and we need to retry or we lost a race to the thread
1004 * unlocking the umtx.
1005 */
1006 umtxq_lock(&uq->uq_key);
1007 if (old == owner)
1008 error = umtxq_sleep(uq, "umtx", timeout == NULL ? NULL :
1009 &timo);
1010 umtxq_remove(uq);
1011 umtxq_unlock(&uq->uq_key);
1012 umtx_key_release(&uq->uq_key);
1013
1014 if (error == 0)
1015 error = thread_check_susp(td, false);
1016 }
1017
1018 if (timeout == NULL) {
1019 /* Mutex locking is restarted if it is interrupted. */
1020 if (error == EINTR)
1021 error = ERESTART;
1022 } else {
1023 /* Timed-locking is not restarted. */
1024 if (error == ERESTART)
1025 error = EINTR;
1026 }
1027 return (error);
1028 }
1029
1030 /*
1031 * Unlock a umtx object.
1032 */
1033 static int
1034 do_unlock_umtx(struct thread *td, struct umtx *umtx, u_long id)
1035 {
1036 struct umtx_key key;
1037 u_long owner;
1038 u_long old;
1039 int error;
1040 int count;
1041
1042 /*
1043 * Make sure we own this mtx.
1044 */
1045 owner = fuword(__DEVOLATILE(u_long *, &umtx->u_owner));
1046 if (owner == -1)
1047 return (EFAULT);
1048
1049 if ((owner & ~UMTX_CONTESTED) != id)
1050 return (EPERM);
1051
1052 /* This should be done in userland */
1053 if ((owner & UMTX_CONTESTED) == 0) {
1054 old = casuword(&umtx->u_owner, owner, UMTX_UNOWNED);
1055 if (old == -1)
1056 return (EFAULT);
1057 if (old == owner)
1058 return (0);
1059 owner = old;
1060 }
1061
1062 /* We should only ever be in here for contested locks */
1063 if ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK, AUTO_SHARE,
1064 &key)) != 0)
1065 return (error);
1066
1067 umtxq_lock(&key);
1068 umtxq_busy(&key);
1069 count = umtxq_count(&key);
1070 umtxq_unlock(&key);
1071
1072 /*
1073 * When unlocking the umtx, it must be marked as unowned if
1074 * there is zero or one thread only waiting for it.
1075 * Otherwise, it must be marked as contested.
1076 */
1077 old = casuword(&umtx->u_owner, owner,
1078 count <= 1 ? UMTX_UNOWNED : UMTX_CONTESTED);
1079 umtxq_lock(&key);
1080 umtxq_signal(&key,1);
1081 umtxq_unbusy(&key);
1082 umtxq_unlock(&key);
1083 umtx_key_release(&key);
1084 if (old == -1)
1085 return (EFAULT);
1086 if (old != owner)
1087 return (EINVAL);
1088 return (0);
1089 }
1090
1091 #ifdef COMPAT_FREEBSD32
1092
1093 /*
1094 * Lock a umtx object.
1095 */
1096 static int
1097 do_lock_umtx32(struct thread *td, uint32_t *m, uint32_t id,
1098 const struct timespec *timeout)
1099 {
1100 struct abs_timeout timo;
1101 struct umtx_q *uq;
1102 uint32_t owner;
1103 uint32_t old;
1104 int error = 0;
1105
1106 uq = td->td_umtxq;
1107
1108 if (timeout != NULL)
1109 abs_timeout_init(&timo, CLOCK_REALTIME, 0, timeout);
1110
1111 /*
1112 * Care must be exercised when dealing with umtx structure. It
1113 * can fault on any access.
1114 */
1115 for (;;) {
1116 /*
1117 * Try the uncontested case. This should be done in userland.
1118 */
1119 owner = casuword32(m, UMUTEX_UNOWNED, id);
1120
1121 /* The acquire succeeded. */
1122 if (owner == UMUTEX_UNOWNED)
1123 return (0);
1124
1125 /* The address was invalid. */
1126 if (owner == -1)
1127 return (EFAULT);
1128
1129 /* If no one owns it but it is contested try to acquire it. */
1130 if (owner == UMUTEX_CONTESTED) {
1131 owner = casuword32(m,
1132 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
1133 if (owner == UMUTEX_CONTESTED)
1134 return (0);
1135
1136 /* The address was invalid. */
1137 if (owner == -1)
1138 return (EFAULT);
1139
1140 error = thread_check_susp(td, false);
1141 if (error != 0)
1142 break;
1143
1144 /* If this failed the lock has changed, restart. */
1145 continue;
1146 }
1147
1148 /*
1149 * If we caught a signal, we have retried and now
1150 * exit immediately.
1151 */
1152 if (error != 0)
1153 return (error);
1154
1155 if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK,
1156 AUTO_SHARE, &uq->uq_key)) != 0)
1157 return (error);
1158
1159 umtxq_lock(&uq->uq_key);
1160 umtxq_busy(&uq->uq_key);
1161 umtxq_insert(uq);
1162 umtxq_unbusy(&uq->uq_key);
1163 umtxq_unlock(&uq->uq_key);
1164
1165 /*
1166 * Set the contested bit so that a release in user space
1167 * knows to use the system call for unlock. If this fails
1168 * either some one else has acquired the lock or it has been
1169 * released.
1170 */
1171 old = casuword32(m, owner, owner | UMUTEX_CONTESTED);
1172
1173 /* The address was invalid. */
1174 if (old == -1) {
1175 umtxq_lock(&uq->uq_key);
1176 umtxq_remove(uq);
1177 umtxq_unlock(&uq->uq_key);
1178 umtx_key_release(&uq->uq_key);
1179 return (EFAULT);
1180 }
1181
1182 /*
1183 * We set the contested bit, sleep. Otherwise the lock changed
1184 * and we need to retry or we lost a race to the thread
1185 * unlocking the umtx.
1186 */
1187 umtxq_lock(&uq->uq_key);
1188 if (old == owner)
1189 error = umtxq_sleep(uq, "umtx", timeout == NULL ?
1190 NULL : &timo);
1191 umtxq_remove(uq);
1192 umtxq_unlock(&uq->uq_key);
1193 umtx_key_release(&uq->uq_key);
1194
1195 if (error == 0)
1196 error = thread_check_susp(td, false);
1197 }
1198
1199 if (timeout == NULL) {
1200 /* Mutex locking is restarted if it is interrupted. */
1201 if (error == EINTR)
1202 error = ERESTART;
1203 } else {
1204 /* Timed-locking is not restarted. */
1205 if (error == ERESTART)
1206 error = EINTR;
1207 }
1208 return (error);
1209 }
1210
1211 /*
1212 * Unlock a umtx object.
1213 */
1214 static int
1215 do_unlock_umtx32(struct thread *td, uint32_t *m, uint32_t id)
1216 {
1217 struct umtx_key key;
1218 uint32_t owner;
1219 uint32_t old;
1220 int error;
1221 int count;
1222
1223 /*
1224 * Make sure we own this mtx.
1225 */
1226 owner = fuword32(m);
1227 if (owner == -1)
1228 return (EFAULT);
1229
1230 if ((owner & ~UMUTEX_CONTESTED) != id)
1231 return (EPERM);
1232
1233 /* This should be done in userland */
1234 if ((owner & UMUTEX_CONTESTED) == 0) {
1235 old = casuword32(m, owner, UMUTEX_UNOWNED);
1236 if (old == -1)
1237 return (EFAULT);
1238 if (old == owner)
1239 return (0);
1240 owner = old;
1241 }
1242
1243 /* We should only ever be in here for contested locks */
1244 if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK, AUTO_SHARE,
1245 &key)) != 0)
1246 return (error);
1247
1248 umtxq_lock(&key);
1249 umtxq_busy(&key);
1250 count = umtxq_count(&key);
1251 umtxq_unlock(&key);
1252
1253 /*
1254 * When unlocking the umtx, it must be marked as unowned if
1255 * there is zero or one thread only waiting for it.
1256 * Otherwise, it must be marked as contested.
1257 */
1258 old = casuword32(m, owner,
1259 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
1260 umtxq_lock(&key);
1261 umtxq_signal(&key,1);
1262 umtxq_unbusy(&key);
1263 umtxq_unlock(&key);
1264 umtx_key_release(&key);
1265 if (old == -1)
1266 return (EFAULT);
1267 if (old != owner)
1268 return (EINVAL);
1269 return (0);
1270 }
1271 #endif /* COMPAT_FREEBSD32 */
1272 #endif /* COMPAT_FREEBSD10 */
1273
1274 /*
1275 * Fetch and compare value, sleep on the address if value is not changed.
1276 */
1277 static int
1278 do_wait(struct thread *td, void *addr, u_long id,
1279 struct _umtx_time *timeout, int compat32, int is_private)
1280 {
1281 struct abs_timeout timo;
1282 struct umtx_q *uq;
1283 u_long tmp;
1284 uint32_t tmp32;
1285 int error = 0;
1286
1287 uq = td->td_umtxq;
1288 if ((error = umtx_key_get(addr, TYPE_SIMPLE_WAIT,
1289 is_private ? THREAD_SHARE : AUTO_SHARE, &uq->uq_key)) != 0)
1290 return (error);
1291
1292 if (timeout != NULL)
1293 abs_timeout_init2(&timo, timeout);
1294
1295 umtxq_lock(&uq->uq_key);
1296 umtxq_insert(uq);
1297 umtxq_unlock(&uq->uq_key);
1298 if (compat32 == 0) {
1299 error = fueword(addr, &tmp);
1300 if (error != 0)
1301 error = EFAULT;
1302 } else {
1303 error = fueword32(addr, &tmp32);
1304 if (error == 0)
1305 tmp = tmp32;
1306 else
1307 error = EFAULT;
1308 }
1309 umtxq_lock(&uq->uq_key);
1310 if (error == 0) {
1311 if (tmp == id)
1312 error = umtxq_sleep(uq, "uwait", timeout == NULL ?
1313 NULL : &timo);
1314 if ((uq->uq_flags & UQF_UMTXQ) == 0)
1315 error = 0;
1316 else
1317 umtxq_remove(uq);
1318 } else if ((uq->uq_flags & UQF_UMTXQ) != 0) {
1319 umtxq_remove(uq);
1320 }
1321 umtxq_unlock(&uq->uq_key);
1322 umtx_key_release(&uq->uq_key);
1323 if (error == ERESTART)
1324 error = EINTR;
1325 return (error);
1326 }
1327
1328 /*
1329 * Wake up threads sleeping on the specified address.
1330 */
1331 int
1332 kern_umtx_wake(struct thread *td, void *uaddr, int n_wake, int is_private)
1333 {
1334 struct umtx_key key;
1335 int ret;
1336
1337 if ((ret = umtx_key_get(uaddr, TYPE_SIMPLE_WAIT,
1338 is_private ? THREAD_SHARE : AUTO_SHARE, &key)) != 0)
1339 return (ret);
1340 umtxq_lock(&key);
1341 umtxq_signal(&key, n_wake);
1342 umtxq_unlock(&key);
1343 umtx_key_release(&key);
1344 return (0);
1345 }
1346
1347 /*
1348 * Lock PTHREAD_PRIO_NONE protocol POSIX mutex.
1349 */
1350 static int
1351 do_lock_normal(struct thread *td, struct umutex *m, uint32_t flags,
1352 struct _umtx_time *timeout, int mode)
1353 {
1354 struct abs_timeout timo;
1355 struct umtx_q *uq;
1356 uint32_t owner, old, id;
1357 int error, rv;
1358
1359 id = td->td_tid;
1360 uq = td->td_umtxq;
1361 error = 0;
1362 if (timeout != NULL)
1363 abs_timeout_init2(&timo, timeout);
1364
1365 /*
1366 * Care must be exercised when dealing with umtx structure. It
1367 * can fault on any access.
1368 */
1369 for (;;) {
1370 rv = fueword32(&m->m_owner, &owner);
1371 if (rv == -1)
1372 return (EFAULT);
1373 if (mode == _UMUTEX_WAIT) {
1374 if (owner == UMUTEX_UNOWNED ||
1375 owner == UMUTEX_CONTESTED ||
1376 owner == UMUTEX_RB_OWNERDEAD ||
1377 owner == UMUTEX_RB_NOTRECOV)
1378 return (0);
1379 } else {
1380 /*
1381 * Robust mutex terminated. Kernel duty is to
1382 * return EOWNERDEAD to the userspace. The
1383 * umutex.m_flags UMUTEX_NONCONSISTENT is set
1384 * by the common userspace code.
1385 */
1386 if (owner == UMUTEX_RB_OWNERDEAD) {
1387 rv = casueword32(&m->m_owner,
1388 UMUTEX_RB_OWNERDEAD, &owner,
1389 id | UMUTEX_CONTESTED);
1390 if (rv == -1)
1391 return (EFAULT);
1392 if (rv == 0) {
1393 MPASS(owner == UMUTEX_RB_OWNERDEAD);
1394 return (EOWNERDEAD); /* success */
1395 }
1396 MPASS(rv == 1);
1397 rv = thread_check_susp(td, false);
1398 if (rv != 0)
1399 return (rv);
1400 continue;
1401 }
1402 if (owner == UMUTEX_RB_NOTRECOV)
1403 return (ENOTRECOVERABLE);
1404
1405 /*
1406 * Try the uncontested case. This should be
1407 * done in userland.
1408 */
1409 rv = casueword32(&m->m_owner, UMUTEX_UNOWNED,
1410 &owner, id);
1411 /* The address was invalid. */
1412 if (rv == -1)
1413 return (EFAULT);
1414
1415 /* The acquire succeeded. */
1416 if (rv == 0) {
1417 MPASS(owner == UMUTEX_UNOWNED);
1418 return (0);
1419 }
1420
1421 /*
1422 * If no one owns it but it is contested try
1423 * to acquire it.
1424 */
1425 MPASS(rv == 1);
1426 if (owner == UMUTEX_CONTESTED) {
1427 rv = casueword32(&m->m_owner,
1428 UMUTEX_CONTESTED, &owner,
1429 id | UMUTEX_CONTESTED);
1430 /* The address was invalid. */
1431 if (rv == -1)
1432 return (EFAULT);
1433 if (rv == 0) {
1434 MPASS(owner == UMUTEX_CONTESTED);
1435 return (0);
1436 }
1437 if (rv == 1) {
1438 rv = thread_check_susp(td, false);
1439 if (rv != 0)
1440 return (rv);
1441 }
1442
1443 /*
1444 * If this failed the lock has
1445 * changed, restart.
1446 */
1447 continue;
1448 }
1449
1450 /* rv == 1 but not contested, likely store failure */
1451 rv = thread_check_susp(td, false);
1452 if (rv != 0)
1453 return (rv);
1454 }
1455
1456 if (mode == _UMUTEX_TRY)
1457 return (EBUSY);
1458
1459 /*
1460 * If we caught a signal, we have retried and now
1461 * exit immediately.
1462 */
1463 if (error != 0)
1464 return (error);
1465
1466 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX,
1467 GET_SHARE(flags), &uq->uq_key)) != 0)
1468 return (error);
1469
1470 umtxq_lock(&uq->uq_key);
1471 umtxq_busy(&uq->uq_key);
1472 umtxq_insert(uq);
1473 umtxq_unlock(&uq->uq_key);
1474
1475 /*
1476 * Set the contested bit so that a release in user space
1477 * knows to use the system call for unlock. If this fails
1478 * either some one else has acquired the lock or it has been
1479 * released.
1480 */
1481 rv = casueword32(&m->m_owner, owner, &old,
1482 owner | UMUTEX_CONTESTED);
1483
1484 /* The address was invalid or casueword failed to store. */
1485 if (rv == -1 || rv == 1) {
1486 umtxq_lock(&uq->uq_key);
1487 umtxq_remove(uq);
1488 umtxq_unbusy(&uq->uq_key);
1489 umtxq_unlock(&uq->uq_key);
1490 umtx_key_release(&uq->uq_key);
1491 if (rv == -1)
1492 return (EFAULT);
1493 if (rv == 1) {
1494 rv = thread_check_susp(td, false);
1495 if (rv != 0)
1496 return (rv);
1497 }
1498 continue;
1499 }
1500
1501 /*
1502 * We set the contested bit, sleep. Otherwise the lock changed
1503 * and we need to retry or we lost a race to the thread
1504 * unlocking the umtx.
1505 */
1506 umtxq_lock(&uq->uq_key);
1507 umtxq_unbusy(&uq->uq_key);
1508 MPASS(old == owner);
1509 error = umtxq_sleep(uq, "umtxn", timeout == NULL ?
1510 NULL : &timo);
1511 umtxq_remove(uq);
1512 umtxq_unlock(&uq->uq_key);
1513 umtx_key_release(&uq->uq_key);
1514
1515 if (error == 0)
1516 error = thread_check_susp(td, false);
1517 }
1518
1519 return (0);
1520 }
1521
1522 /*
1523 * Unlock PTHREAD_PRIO_NONE protocol POSIX mutex.
1524 */
1525 static int
1526 do_unlock_normal(struct thread *td, struct umutex *m, uint32_t flags, bool rb)
1527 {
1528 struct umtx_key key;
1529 uint32_t owner, old, id, newlock;
1530 int error, count;
1531
1532 id = td->td_tid;
1533
1534 again:
1535 /*
1536 * Make sure we own this mtx.
1537 */
1538 error = fueword32(&m->m_owner, &owner);
1539 if (error == -1)
1540 return (EFAULT);
1541
1542 if ((owner & ~UMUTEX_CONTESTED) != id)
1543 return (EPERM);
1544
1545 newlock = umtx_unlock_val(flags, rb);
1546 if ((owner & UMUTEX_CONTESTED) == 0) {
1547 error = casueword32(&m->m_owner, owner, &old, newlock);
1548 if (error == -1)
1549 return (EFAULT);
1550 if (error == 1) {
1551 error = thread_check_susp(td, false);
1552 if (error != 0)
1553 return (error);
1554 goto again;
1555 }
1556 MPASS(old == owner);
1557 return (0);
1558 }
1559
1560 /* We should only ever be in here for contested locks */
1561 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags),
1562 &key)) != 0)
1563 return (error);
1564
1565 umtxq_lock(&key);
1566 umtxq_busy(&key);
1567 count = umtxq_count(&key);
1568 umtxq_unlock(&key);
1569
1570 /*
1571 * When unlocking the umtx, it must be marked as unowned if
1572 * there is zero or one thread only waiting for it.
1573 * Otherwise, it must be marked as contested.
1574 */
1575 if (count > 1)
1576 newlock |= UMUTEX_CONTESTED;
1577 error = casueword32(&m->m_owner, owner, &old, newlock);
1578 umtxq_lock(&key);
1579 umtxq_signal(&key, 1);
1580 umtxq_unbusy(&key);
1581 umtxq_unlock(&key);
1582 umtx_key_release(&key);
1583 if (error == -1)
1584 return (EFAULT);
1585 if (error == 1) {
1586 if (old != owner)
1587 return (EINVAL);
1588 error = thread_check_susp(td, false);
1589 if (error != 0)
1590 return (error);
1591 goto again;
1592 }
1593 return (0);
1594 }
1595
1596 /*
1597 * Check if the mutex is available and wake up a waiter,
1598 * only for simple mutex.
1599 */
1600 static int
1601 do_wake_umutex(struct thread *td, struct umutex *m)
1602 {
1603 struct umtx_key key;
1604 uint32_t owner;
1605 uint32_t flags;
1606 int error;
1607 int count;
1608
1609 again:
1610 error = fueword32(&m->m_owner, &owner);
1611 if (error == -1)
1612 return (EFAULT);
1613
1614 if ((owner & ~UMUTEX_CONTESTED) != 0 && owner != UMUTEX_RB_OWNERDEAD &&
1615 owner != UMUTEX_RB_NOTRECOV)
1616 return (0);
1617
1618 error = fueword32(&m->m_flags, &flags);
1619 if (error == -1)
1620 return (EFAULT);
1621
1622 /* We should only ever be in here for contested locks */
1623 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags),
1624 &key)) != 0)
1625 return (error);
1626
1627 umtxq_lock(&key);
1628 umtxq_busy(&key);
1629 count = umtxq_count(&key);
1630 umtxq_unlock(&key);
1631
1632 if (count <= 1 && owner != UMUTEX_RB_OWNERDEAD &&
1633 owner != UMUTEX_RB_NOTRECOV) {
1634 error = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
1635 UMUTEX_UNOWNED);
1636 if (error == -1) {
1637 error = EFAULT;
1638 } else if (error == 1) {
1639 umtxq_lock(&key);
1640 umtxq_unbusy(&key);
1641 umtxq_unlock(&key);
1642 umtx_key_release(&key);
1643 error = thread_check_susp(td, false);
1644 if (error != 0)
1645 return (error);
1646 goto again;
1647 }
1648 }
1649
1650 umtxq_lock(&key);
1651 if (error == 0 && count != 0) {
1652 MPASS((owner & ~UMUTEX_CONTESTED) == 0 ||
1653 owner == UMUTEX_RB_OWNERDEAD ||
1654 owner == UMUTEX_RB_NOTRECOV);
1655 umtxq_signal(&key, 1);
1656 }
1657 umtxq_unbusy(&key);
1658 umtxq_unlock(&key);
1659 umtx_key_release(&key);
1660 return (error);
1661 }
1662
1663 /*
1664 * Check if the mutex has waiters and tries to fix contention bit.
1665 */
1666 static int
1667 do_wake2_umutex(struct thread *td, struct umutex *m, uint32_t flags)
1668 {
1669 struct umtx_key key;
1670 uint32_t owner, old;
1671 int type;
1672 int error;
1673 int count;
1674
1675 switch (flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT |
1676 UMUTEX_ROBUST)) {
1677 case 0:
1678 case UMUTEX_ROBUST:
1679 type = TYPE_NORMAL_UMUTEX;
1680 break;
1681 case UMUTEX_PRIO_INHERIT:
1682 type = TYPE_PI_UMUTEX;
1683 break;
1684 case (UMUTEX_PRIO_INHERIT | UMUTEX_ROBUST):
1685 type = TYPE_PI_ROBUST_UMUTEX;
1686 break;
1687 case UMUTEX_PRIO_PROTECT:
1688 type = TYPE_PP_UMUTEX;
1689 break;
1690 case (UMUTEX_PRIO_PROTECT | UMUTEX_ROBUST):
1691 type = TYPE_PP_ROBUST_UMUTEX;
1692 break;
1693 default:
1694 return (EINVAL);
1695 }
1696 if ((error = umtx_key_get(m, type, GET_SHARE(flags), &key)) != 0)
1697 return (error);
1698
1699 owner = 0;
1700 umtxq_lock(&key);
1701 umtxq_busy(&key);
1702 count = umtxq_count(&key);
1703 umtxq_unlock(&key);
1704
1705 error = fueword32(&m->m_owner, &owner);
1706 if (error == -1)
1707 error = EFAULT;
1708
1709 /*
1710 * Only repair contention bit if there is a waiter, this means
1711 * the mutex is still being referenced by userland code,
1712 * otherwise don't update any memory.
1713 */
1714 while (error == 0 && (owner & UMUTEX_CONTESTED) == 0 &&
1715 (count > 1 || (count == 1 && (owner & ~UMUTEX_CONTESTED) != 0))) {
1716 error = casueword32(&m->m_owner, owner, &old,
1717 owner | UMUTEX_CONTESTED);
1718 if (error == -1) {
1719 error = EFAULT;
1720 break;
1721 }
1722 if (error == 0) {
1723 MPASS(old == owner);
1724 break;
1725 }
1726 owner = old;
1727 error = thread_check_susp(td, false);
1728 }
1729
1730 umtxq_lock(&key);
1731 if (error == EFAULT) {
1732 umtxq_signal(&key, INT_MAX);
1733 } else if (count != 0 && ((owner & ~UMUTEX_CONTESTED) == 0 ||
1734 owner == UMUTEX_RB_OWNERDEAD || owner == UMUTEX_RB_NOTRECOV))
1735 umtxq_signal(&key, 1);
1736 umtxq_unbusy(&key);
1737 umtxq_unlock(&key);
1738 umtx_key_release(&key);
1739 return (error);
1740 }
1741
1742 static inline struct umtx_pi *
1743 umtx_pi_alloc(int flags)
1744 {
1745 struct umtx_pi *pi;
1746
1747 pi = uma_zalloc(umtx_pi_zone, M_ZERO | flags);
1748 TAILQ_INIT(&pi->pi_blocked);
1749 atomic_add_int(&umtx_pi_allocated, 1);
1750 return (pi);
1751 }
1752
1753 static inline void
1754 umtx_pi_free(struct umtx_pi *pi)
1755 {
1756 uma_zfree(umtx_pi_zone, pi);
1757 atomic_add_int(&umtx_pi_allocated, -1);
1758 }
1759
1760 /*
1761 * Adjust the thread's position on a pi_state after its priority has been
1762 * changed.
1763 */
1764 static int
1765 umtx_pi_adjust_thread(struct umtx_pi *pi, struct thread *td)
1766 {
1767 struct umtx_q *uq, *uq1, *uq2;
1768 struct thread *td1;
1769
1770 mtx_assert(&umtx_lock, MA_OWNED);
1771 if (pi == NULL)
1772 return (0);
1773
1774 uq = td->td_umtxq;
1775
1776 /*
1777 * Check if the thread needs to be moved on the blocked chain.
1778 * It needs to be moved if either its priority is lower than
1779 * the previous thread or higher than the next thread.
1780 */
1781 uq1 = TAILQ_PREV(uq, umtxq_head, uq_lockq);
1782 uq2 = TAILQ_NEXT(uq, uq_lockq);
1783 if ((uq1 != NULL && UPRI(td) < UPRI(uq1->uq_thread)) ||
1784 (uq2 != NULL && UPRI(td) > UPRI(uq2->uq_thread))) {
1785 /*
1786 * Remove thread from blocked chain and determine where
1787 * it should be moved to.
1788 */
1789 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq);
1790 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) {
1791 td1 = uq1->uq_thread;
1792 MPASS(td1->td_proc->p_magic == P_MAGIC);
1793 if (UPRI(td1) > UPRI(td))
1794 break;
1795 }
1796
1797 if (uq1 == NULL)
1798 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq);
1799 else
1800 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq);
1801 }
1802 return (1);
1803 }
1804
1805 static struct umtx_pi *
1806 umtx_pi_next(struct umtx_pi *pi)
1807 {
1808 struct umtx_q *uq_owner;
1809
1810 if (pi->pi_owner == NULL)
1811 return (NULL);
1812 uq_owner = pi->pi_owner->td_umtxq;
1813 if (uq_owner == NULL)
1814 return (NULL);
1815 return (uq_owner->uq_pi_blocked);
1816 }
1817
1818 /*
1819 * Floyd's Cycle-Finding Algorithm.
1820 */
1821 static bool
1822 umtx_pi_check_loop(struct umtx_pi *pi)
1823 {
1824 struct umtx_pi *pi1; /* fast iterator */
1825
1826 mtx_assert(&umtx_lock, MA_OWNED);
1827 if (pi == NULL)
1828 return (false);
1829 pi1 = pi;
1830 for (;;) {
1831 pi = umtx_pi_next(pi);
1832 if (pi == NULL)
1833 break;
1834 pi1 = umtx_pi_next(pi1);
1835 if (pi1 == NULL)
1836 break;
1837 pi1 = umtx_pi_next(pi1);
1838 if (pi1 == NULL)
1839 break;
1840 if (pi == pi1)
1841 return (true);
1842 }
1843 return (false);
1844 }
1845
1846 /*
1847 * Propagate priority when a thread is blocked on POSIX
1848 * PI mutex.
1849 */
1850 static void
1851 umtx_propagate_priority(struct thread *td)
1852 {
1853 struct umtx_q *uq;
1854 struct umtx_pi *pi;
1855 int pri;
1856
1857 mtx_assert(&umtx_lock, MA_OWNED);
1858 pri = UPRI(td);
1859 uq = td->td_umtxq;
1860 pi = uq->uq_pi_blocked;
1861 if (pi == NULL)
1862 return;
1863 if (umtx_pi_check_loop(pi))
1864 return;
1865
1866 for (;;) {
1867 td = pi->pi_owner;
1868 if (td == NULL || td == curthread)
1869 return;
1870
1871 MPASS(td->td_proc != NULL);
1872 MPASS(td->td_proc->p_magic == P_MAGIC);
1873
1874 thread_lock(td);
1875 if (td->td_lend_user_pri > pri)
1876 sched_lend_user_prio(td, pri);
1877 else {
1878 thread_unlock(td);
1879 break;
1880 }
1881 thread_unlock(td);
1882
1883 /*
1884 * Pick up the lock that td is blocked on.
1885 */
1886 uq = td->td_umtxq;
1887 pi = uq->uq_pi_blocked;
1888 if (pi == NULL)
1889 break;
1890 /* Resort td on the list if needed. */
1891 umtx_pi_adjust_thread(pi, td);
1892 }
1893 }
1894
1895 /*
1896 * Unpropagate priority for a PI mutex when a thread blocked on
1897 * it is interrupted by signal or resumed by others.
1898 */
1899 static void
1900 umtx_repropagate_priority(struct umtx_pi *pi)
1901 {
1902 struct umtx_q *uq, *uq_owner;
1903 struct umtx_pi *pi2;
1904 int pri;
1905
1906 mtx_assert(&umtx_lock, MA_OWNED);
1907
1908 if (umtx_pi_check_loop(pi))
1909 return;
1910 while (pi != NULL && pi->pi_owner != NULL) {
1911 pri = PRI_MAX;
1912 uq_owner = pi->pi_owner->td_umtxq;
1913
1914 TAILQ_FOREACH(pi2, &uq_owner->uq_pi_contested, pi_link) {
1915 uq = TAILQ_FIRST(&pi2->pi_blocked);
1916 if (uq != NULL) {
1917 if (pri > UPRI(uq->uq_thread))
1918 pri = UPRI(uq->uq_thread);
1919 }
1920 }
1921
1922 if (pri > uq_owner->uq_inherited_pri)
1923 pri = uq_owner->uq_inherited_pri;
1924 thread_lock(pi->pi_owner);
1925 sched_lend_user_prio(pi->pi_owner, pri);
1926 thread_unlock(pi->pi_owner);
1927 if ((pi = uq_owner->uq_pi_blocked) != NULL)
1928 umtx_pi_adjust_thread(pi, uq_owner->uq_thread);
1929 }
1930 }
1931
1932 /*
1933 * Insert a PI mutex into owned list.
1934 */
1935 static void
1936 umtx_pi_setowner(struct umtx_pi *pi, struct thread *owner)
1937 {
1938 struct umtx_q *uq_owner;
1939
1940 uq_owner = owner->td_umtxq;
1941 mtx_assert(&umtx_lock, MA_OWNED);
1942 MPASS(pi->pi_owner == NULL);
1943 pi->pi_owner = owner;
1944 TAILQ_INSERT_TAIL(&uq_owner->uq_pi_contested, pi, pi_link);
1945 }
1946
1947 /*
1948 * Disown a PI mutex, and remove it from the owned list.
1949 */
1950 static void
1951 umtx_pi_disown(struct umtx_pi *pi)
1952 {
1953
1954 mtx_assert(&umtx_lock, MA_OWNED);
1955 TAILQ_REMOVE(&pi->pi_owner->td_umtxq->uq_pi_contested, pi, pi_link);
1956 pi->pi_owner = NULL;
1957 }
1958
1959 /*
1960 * Claim ownership of a PI mutex.
1961 */
1962 static int
1963 umtx_pi_claim(struct umtx_pi *pi, struct thread *owner)
1964 {
1965 struct umtx_q *uq;
1966 int pri;
1967
1968 mtx_lock(&umtx_lock);
1969 if (pi->pi_owner == owner) {
1970 mtx_unlock(&umtx_lock);
1971 return (0);
1972 }
1973
1974 if (pi->pi_owner != NULL) {
1975 /*
1976 * userland may have already messed the mutex, sigh.
1977 */
1978 mtx_unlock(&umtx_lock);
1979 return (EPERM);
1980 }
1981 umtx_pi_setowner(pi, owner);
1982 uq = TAILQ_FIRST(&pi->pi_blocked);
1983 if (uq != NULL) {
1984 pri = UPRI(uq->uq_thread);
1985 thread_lock(owner);
1986 if (pri < UPRI(owner))
1987 sched_lend_user_prio(owner, pri);
1988 thread_unlock(owner);
1989 }
1990 mtx_unlock(&umtx_lock);
1991 return (0);
1992 }
1993
1994 /*
1995 * Adjust a thread's order position in its blocked PI mutex,
1996 * this may result new priority propagating process.
1997 */
1998 void
1999 umtx_pi_adjust(struct thread *td, u_char oldpri)
2000 {
2001 struct umtx_q *uq;
2002 struct umtx_pi *pi;
2003
2004 uq = td->td_umtxq;
2005 mtx_lock(&umtx_lock);
2006 /*
2007 * Pick up the lock that td is blocked on.
2008 */
2009 pi = uq->uq_pi_blocked;
2010 if (pi != NULL) {
2011 umtx_pi_adjust_thread(pi, td);
2012 umtx_repropagate_priority(pi);
2013 }
2014 mtx_unlock(&umtx_lock);
2015 }
2016
2017 /*
2018 * Sleep on a PI mutex.
2019 */
2020 static int
2021 umtxq_sleep_pi(struct umtx_q *uq, struct umtx_pi *pi, uint32_t owner,
2022 const char *wmesg, struct abs_timeout *timo, bool shared)
2023 {
2024 struct thread *td, *td1;
2025 struct umtx_q *uq1;
2026 int error, pri;
2027 #ifdef INVARIANTS
2028 struct umtxq_chain *uc;
2029
2030 uc = umtxq_getchain(&pi->pi_key);
2031 #endif
2032 error = 0;
2033 td = uq->uq_thread;
2034 KASSERT(td == curthread, ("inconsistent uq_thread"));
2035 UMTXQ_LOCKED_ASSERT(umtxq_getchain(&uq->uq_key));
2036 KASSERT(uc->uc_busy != 0, ("umtx chain is not busy"));
2037 umtxq_insert(uq);
2038 mtx_lock(&umtx_lock);
2039 if (pi->pi_owner == NULL) {
2040 mtx_unlock(&umtx_lock);
2041 td1 = tdfind(owner, shared ? -1 : td->td_proc->p_pid);
2042 mtx_lock(&umtx_lock);
2043 if (td1 != NULL) {
2044 if (pi->pi_owner == NULL)
2045 umtx_pi_setowner(pi, td1);
2046 PROC_UNLOCK(td1->td_proc);
2047 }
2048 }
2049
2050 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) {
2051 pri = UPRI(uq1->uq_thread);
2052 if (pri > UPRI(td))
2053 break;
2054 }
2055
2056 if (uq1 != NULL)
2057 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq);
2058 else
2059 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq);
2060
2061 uq->uq_pi_blocked = pi;
2062 thread_lock(td);
2063 td->td_flags |= TDF_UPIBLOCKED;
2064 thread_unlock(td);
2065 umtx_propagate_priority(td);
2066 mtx_unlock(&umtx_lock);
2067 umtxq_unbusy(&uq->uq_key);
2068
2069 error = umtxq_sleep(uq, wmesg, timo);
2070 umtxq_remove(uq);
2071
2072 mtx_lock(&umtx_lock);
2073 uq->uq_pi_blocked = NULL;
2074 thread_lock(td);
2075 td->td_flags &= ~TDF_UPIBLOCKED;
2076 thread_unlock(td);
2077 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq);
2078 umtx_repropagate_priority(pi);
2079 mtx_unlock(&umtx_lock);
2080 umtxq_unlock(&uq->uq_key);
2081
2082 return (error);
2083 }
2084
2085 /*
2086 * Add reference count for a PI mutex.
2087 */
2088 static void
2089 umtx_pi_ref(struct umtx_pi *pi)
2090 {
2091
2092 UMTXQ_LOCKED_ASSERT(umtxq_getchain(&pi->pi_key));
2093 pi->pi_refcount++;
2094 }
2095
2096 /*
2097 * Decrease reference count for a PI mutex, if the counter
2098 * is decreased to zero, its memory space is freed.
2099 */
2100 static void
2101 umtx_pi_unref(struct umtx_pi *pi)
2102 {
2103 struct umtxq_chain *uc;
2104
2105 uc = umtxq_getchain(&pi->pi_key);
2106 UMTXQ_LOCKED_ASSERT(uc);
2107 KASSERT(pi->pi_refcount > 0, ("invalid reference count"));
2108 if (--pi->pi_refcount == 0) {
2109 mtx_lock(&umtx_lock);
2110 if (pi->pi_owner != NULL)
2111 umtx_pi_disown(pi);
2112 KASSERT(TAILQ_EMPTY(&pi->pi_blocked),
2113 ("blocked queue not empty"));
2114 mtx_unlock(&umtx_lock);
2115 TAILQ_REMOVE(&uc->uc_pi_list, pi, pi_hashlink);
2116 umtx_pi_free(pi);
2117 }
2118 }
2119
2120 /*
2121 * Find a PI mutex in hash table.
2122 */
2123 static struct umtx_pi *
2124 umtx_pi_lookup(struct umtx_key *key)
2125 {
2126 struct umtxq_chain *uc;
2127 struct umtx_pi *pi;
2128
2129 uc = umtxq_getchain(key);
2130 UMTXQ_LOCKED_ASSERT(uc);
2131
2132 TAILQ_FOREACH(pi, &uc->uc_pi_list, pi_hashlink) {
2133 if (umtx_key_match(&pi->pi_key, key)) {
2134 return (pi);
2135 }
2136 }
2137 return (NULL);
2138 }
2139
2140 /*
2141 * Insert a PI mutex into hash table.
2142 */
2143 static inline void
2144 umtx_pi_insert(struct umtx_pi *pi)
2145 {
2146 struct umtxq_chain *uc;
2147
2148 uc = umtxq_getchain(&pi->pi_key);
2149 UMTXQ_LOCKED_ASSERT(uc);
2150 TAILQ_INSERT_TAIL(&uc->uc_pi_list, pi, pi_hashlink);
2151 }
2152
2153 /*
2154 * Lock a PI mutex.
2155 */
2156 static int
2157 do_lock_pi(struct thread *td, struct umutex *m, uint32_t flags,
2158 struct _umtx_time *timeout, int try)
2159 {
2160 struct abs_timeout timo;
2161 struct umtx_q *uq;
2162 struct umtx_pi *pi, *new_pi;
2163 uint32_t id, old_owner, owner, old;
2164 int error, rv;
2165
2166 id = td->td_tid;
2167 uq = td->td_umtxq;
2168
2169 if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
2170 TYPE_PI_ROBUST_UMUTEX : TYPE_PI_UMUTEX, GET_SHARE(flags),
2171 &uq->uq_key)) != 0)
2172 return (error);
2173
2174 if (timeout != NULL)
2175 abs_timeout_init2(&timo, timeout);
2176
2177 umtxq_lock(&uq->uq_key);
2178 pi = umtx_pi_lookup(&uq->uq_key);
2179 if (pi == NULL) {
2180 new_pi = umtx_pi_alloc(M_NOWAIT);
2181 if (new_pi == NULL) {
2182 umtxq_unlock(&uq->uq_key);
2183 new_pi = umtx_pi_alloc(M_WAITOK);
2184 umtxq_lock(&uq->uq_key);
2185 pi = umtx_pi_lookup(&uq->uq_key);
2186 if (pi != NULL) {
2187 umtx_pi_free(new_pi);
2188 new_pi = NULL;
2189 }
2190 }
2191 if (new_pi != NULL) {
2192 new_pi->pi_key = uq->uq_key;
2193 umtx_pi_insert(new_pi);
2194 pi = new_pi;
2195 }
2196 }
2197 umtx_pi_ref(pi);
2198 umtxq_unlock(&uq->uq_key);
2199
2200 /*
2201 * Care must be exercised when dealing with umtx structure. It
2202 * can fault on any access.
2203 */
2204 for (;;) {
2205 /*
2206 * Try the uncontested case. This should be done in userland.
2207 */
2208 rv = casueword32(&m->m_owner, UMUTEX_UNOWNED, &owner, id);
2209 /* The address was invalid. */
2210 if (rv == -1) {
2211 error = EFAULT;
2212 break;
2213 }
2214 /* The acquire succeeded. */
2215 if (rv == 0) {
2216 MPASS(owner == UMUTEX_UNOWNED);
2217 error = 0;
2218 break;
2219 }
2220
2221 if (owner == UMUTEX_RB_NOTRECOV) {
2222 error = ENOTRECOVERABLE;
2223 break;
2224 }
2225
2226 /*
2227 * Avoid overwriting a possible error from sleep due
2228 * to the pending signal with suspension check result.
2229 */
2230 if (error == 0) {
2231 error = thread_check_susp(td, true);
2232 if (error != 0)
2233 break;
2234 }
2235
2236 /* If no one owns it but it is contested try to acquire it. */
2237 if (owner == UMUTEX_CONTESTED || owner == UMUTEX_RB_OWNERDEAD) {
2238 old_owner = owner;
2239 rv = casueword32(&m->m_owner, owner, &owner,
2240 id | UMUTEX_CONTESTED);
2241 /* The address was invalid. */
2242 if (rv == -1) {
2243 error = EFAULT;
2244 break;
2245 }
2246 if (rv == 1) {
2247 if (error == 0) {
2248 error = thread_check_susp(td, true);
2249 if (error != 0)
2250 break;
2251 }
2252
2253 /*
2254 * If this failed the lock could
2255 * changed, restart.
2256 */
2257 continue;
2258 }
2259
2260 MPASS(rv == 0);
2261 MPASS(owner == old_owner);
2262 umtxq_lock(&uq->uq_key);
2263 umtxq_busy(&uq->uq_key);
2264 error = umtx_pi_claim(pi, td);
2265 umtxq_unbusy(&uq->uq_key);
2266 umtxq_unlock(&uq->uq_key);
2267 if (error != 0) {
2268 /*
2269 * Since we're going to return an
2270 * error, restore the m_owner to its
2271 * previous, unowned state to avoid
2272 * compounding the problem.
2273 */
2274 (void)casuword32(&m->m_owner,
2275 id | UMUTEX_CONTESTED, old_owner);
2276 }
2277 if (error == 0 && old_owner == UMUTEX_RB_OWNERDEAD)
2278 error = EOWNERDEAD;
2279 break;
2280 }
2281
2282 if ((owner & ~UMUTEX_CONTESTED) == id) {
2283 error = EDEADLK;
2284 break;
2285 }
2286
2287 if (try != 0) {
2288 error = EBUSY;
2289 break;
2290 }
2291
2292 /*
2293 * If we caught a signal, we have retried and now
2294 * exit immediately.
2295 */
2296 if (error != 0)
2297 break;
2298
2299 umtxq_lock(&uq->uq_key);
2300 umtxq_busy(&uq->uq_key);
2301 umtxq_unlock(&uq->uq_key);
2302
2303 /*
2304 * Set the contested bit so that a release in user space
2305 * knows to use the system call for unlock. If this fails
2306 * either some one else has acquired the lock or it has been
2307 * released.
2308 */
2309 rv = casueword32(&m->m_owner, owner, &old, owner |
2310 UMUTEX_CONTESTED);
2311
2312 /* The address was invalid. */
2313 if (rv == -1) {
2314 umtxq_unbusy_unlocked(&uq->uq_key);
2315 error = EFAULT;
2316 break;
2317 }
2318 if (rv == 1) {
2319 umtxq_unbusy_unlocked(&uq->uq_key);
2320 error = thread_check_susp(td, true);
2321 if (error != 0)
2322 break;
2323
2324 /*
2325 * The lock changed and we need to retry or we
2326 * lost a race to the thread unlocking the
2327 * umtx. Note that the UMUTEX_RB_OWNERDEAD
2328 * value for owner is impossible there.
2329 */
2330 continue;
2331 }
2332
2333 umtxq_lock(&uq->uq_key);
2334
2335 /* We set the contested bit, sleep. */
2336 MPASS(old == owner);
2337 error = umtxq_sleep_pi(uq, pi, owner & ~UMUTEX_CONTESTED,
2338 "umtxpi", timeout == NULL ? NULL : &timo,
2339 (flags & USYNC_PROCESS_SHARED) != 0);
2340 if (error != 0)
2341 continue;
2342
2343 error = thread_check_susp(td, false);
2344 if (error != 0)
2345 break;
2346 }
2347
2348 umtxq_lock(&uq->uq_key);
2349 umtx_pi_unref(pi);
2350 umtxq_unlock(&uq->uq_key);
2351
2352 umtx_key_release(&uq->uq_key);
2353 return (error);
2354 }
2355
2356 /*
2357 * Unlock a PI mutex.
2358 */
2359 static int
2360 do_unlock_pi(struct thread *td, struct umutex *m, uint32_t flags, bool rb)
2361 {
2362 struct umtx_key key;
2363 struct umtx_q *uq_first, *uq_first2, *uq_me;
2364 struct umtx_pi *pi, *pi2;
2365 uint32_t id, new_owner, old, owner;
2366 int count, error, pri;
2367
2368 id = td->td_tid;
2369
2370 usrloop:
2371 /*
2372 * Make sure we own this mtx.
2373 */
2374 error = fueword32(&m->m_owner, &owner);
2375 if (error == -1)
2376 return (EFAULT);
2377
2378 if ((owner & ~UMUTEX_CONTESTED) != id)
2379 return (EPERM);
2380
2381 new_owner = umtx_unlock_val(flags, rb);
2382
2383 /* This should be done in userland */
2384 if ((owner & UMUTEX_CONTESTED) == 0) {
2385 error = casueword32(&m->m_owner, owner, &old, new_owner);
2386 if (error == -1)
2387 return (EFAULT);
2388 if (error == 1) {
2389 error = thread_check_susp(td, true);
2390 if (error != 0)
2391 return (error);
2392 goto usrloop;
2393 }
2394 if (old == owner)
2395 return (0);
2396 owner = old;
2397 }
2398
2399 /* We should only ever be in here for contested locks */
2400 if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
2401 TYPE_PI_ROBUST_UMUTEX : TYPE_PI_UMUTEX, GET_SHARE(flags),
2402 &key)) != 0)
2403 return (error);
2404
2405 umtxq_lock(&key);
2406 umtxq_busy(&key);
2407 count = umtxq_count_pi(&key, &uq_first);
2408 if (uq_first != NULL) {
2409 mtx_lock(&umtx_lock);
2410 pi = uq_first->uq_pi_blocked;
2411 KASSERT(pi != NULL, ("pi == NULL?"));
2412 if (pi->pi_owner != td && !(rb && pi->pi_owner == NULL)) {
2413 mtx_unlock(&umtx_lock);
2414 umtxq_unbusy(&key);
2415 umtxq_unlock(&key);
2416 umtx_key_release(&key);
2417 /* userland messed the mutex */
2418 return (EPERM);
2419 }
2420 uq_me = td->td_umtxq;
2421 if (pi->pi_owner == td)
2422 umtx_pi_disown(pi);
2423 /* get highest priority thread which is still sleeping. */
2424 uq_first = TAILQ_FIRST(&pi->pi_blocked);
2425 while (uq_first != NULL &&
2426 (uq_first->uq_flags & UQF_UMTXQ) == 0) {
2427 uq_first = TAILQ_NEXT(uq_first, uq_lockq);
2428 }
2429 pri = PRI_MAX;
2430 TAILQ_FOREACH(pi2, &uq_me->uq_pi_contested, pi_link) {
2431 uq_first2 = TAILQ_FIRST(&pi2->pi_blocked);
2432 if (uq_first2 != NULL) {
2433 if (pri > UPRI(uq_first2->uq_thread))
2434 pri = UPRI(uq_first2->uq_thread);
2435 }
2436 }
2437 thread_lock(td);
2438 sched_lend_user_prio(td, pri);
2439 thread_unlock(td);
2440 mtx_unlock(&umtx_lock);
2441 if (uq_first)
2442 umtxq_signal_thread(uq_first);
2443 } else {
2444 pi = umtx_pi_lookup(&key);
2445 /*
2446 * A umtx_pi can exist if a signal or timeout removed the
2447 * last waiter from the umtxq, but there is still
2448 * a thread in do_lock_pi() holding the umtx_pi.
2449 */
2450 if (pi != NULL) {
2451 /*
2452 * The umtx_pi can be unowned, such as when a thread
2453 * has just entered do_lock_pi(), allocated the
2454 * umtx_pi, and unlocked the umtxq.
2455 * If the current thread owns it, it must disown it.
2456 */
2457 mtx_lock(&umtx_lock);
2458 if (pi->pi_owner == td)
2459 umtx_pi_disown(pi);
2460 mtx_unlock(&umtx_lock);
2461 }
2462 }
2463 umtxq_unlock(&key);
2464
2465 /*
2466 * When unlocking the umtx, it must be marked as unowned if
2467 * there is zero or one thread only waiting for it.
2468 * Otherwise, it must be marked as contested.
2469 */
2470
2471 if (count > 1)
2472 new_owner |= UMUTEX_CONTESTED;
2473 again:
2474 error = casueword32(&m->m_owner, owner, &old, new_owner);
2475 if (error == 1) {
2476 error = thread_check_susp(td, false);
2477 if (error == 0)
2478 goto again;
2479 }
2480 umtxq_unbusy_unlocked(&key);
2481 umtx_key_release(&key);
2482 if (error == -1)
2483 return (EFAULT);
2484 if (error == 0 && old != owner)
2485 return (EINVAL);
2486 return (error);
2487 }
2488
2489 /*
2490 * Lock a PP mutex.
2491 */
2492 static int
2493 do_lock_pp(struct thread *td, struct umutex *m, uint32_t flags,
2494 struct _umtx_time *timeout, int try)
2495 {
2496 struct abs_timeout timo;
2497 struct umtx_q *uq, *uq2;
2498 struct umtx_pi *pi;
2499 uint32_t ceiling;
2500 uint32_t owner, id;
2501 int error, pri, old_inherited_pri, su, rv;
2502
2503 id = td->td_tid;
2504 uq = td->td_umtxq;
2505 if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
2506 TYPE_PP_ROBUST_UMUTEX : TYPE_PP_UMUTEX, GET_SHARE(flags),
2507 &uq->uq_key)) != 0)
2508 return (error);
2509
2510 if (timeout != NULL)
2511 abs_timeout_init2(&timo, timeout);
2512
2513 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0);
2514 for (;;) {
2515 old_inherited_pri = uq->uq_inherited_pri;
2516 umtxq_lock(&uq->uq_key);
2517 umtxq_busy(&uq->uq_key);
2518 umtxq_unlock(&uq->uq_key);
2519
2520 rv = fueword32(&m->m_ceilings[0], &ceiling);
2521 if (rv == -1) {
2522 error = EFAULT;
2523 goto out;
2524 }
2525 ceiling = RTP_PRIO_MAX - ceiling;
2526 if (ceiling > RTP_PRIO_MAX) {
2527 error = EINVAL;
2528 goto out;
2529 }
2530
2531 mtx_lock(&umtx_lock);
2532 if (UPRI(td) < PRI_MIN_REALTIME + ceiling) {
2533 mtx_unlock(&umtx_lock);
2534 error = EINVAL;
2535 goto out;
2536 }
2537 if (su && PRI_MIN_REALTIME + ceiling < uq->uq_inherited_pri) {
2538 uq->uq_inherited_pri = PRI_MIN_REALTIME + ceiling;
2539 thread_lock(td);
2540 if (uq->uq_inherited_pri < UPRI(td))
2541 sched_lend_user_prio(td, uq->uq_inherited_pri);
2542 thread_unlock(td);
2543 }
2544 mtx_unlock(&umtx_lock);
2545
2546 rv = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
2547 id | UMUTEX_CONTESTED);
2548 /* The address was invalid. */
2549 if (rv == -1) {
2550 error = EFAULT;
2551 break;
2552 }
2553 if (rv == 0) {
2554 MPASS(owner == UMUTEX_CONTESTED);
2555 error = 0;
2556 break;
2557 }
2558 /* rv == 1 */
2559 if (owner == UMUTEX_RB_OWNERDEAD) {
2560 rv = casueword32(&m->m_owner, UMUTEX_RB_OWNERDEAD,
2561 &owner, id | UMUTEX_CONTESTED);
2562 if (rv == -1) {
2563 error = EFAULT;
2564 break;
2565 }
2566 if (rv == 0) {
2567 MPASS(owner == UMUTEX_RB_OWNERDEAD);
2568 error = EOWNERDEAD; /* success */
2569 break;
2570 }
2571
2572 /*
2573 * rv == 1, only check for suspension if we
2574 * did not already catched a signal. If we
2575 * get an error from the check, the same
2576 * condition is checked by the umtxq_sleep()
2577 * call below, so we should obliterate the
2578 * error to not skip the last loop iteration.
2579 */
2580 if (error == 0) {
2581 error = thread_check_susp(td, false);
2582 if (error == 0) {
2583 if (try != 0)
2584 error = EBUSY;
2585 else
2586 continue;
2587 }
2588 error = 0;
2589 }
2590 } else if (owner == UMUTEX_RB_NOTRECOV) {
2591 error = ENOTRECOVERABLE;
2592 }
2593
2594 if (try != 0)
2595 error = EBUSY;
2596
2597 /*
2598 * If we caught a signal, we have retried and now
2599 * exit immediately.
2600 */
2601 if (error != 0)
2602 break;
2603
2604 umtxq_lock(&uq->uq_key);
2605 umtxq_insert(uq);
2606 umtxq_unbusy(&uq->uq_key);
2607 error = umtxq_sleep(uq, "umtxpp", timeout == NULL ?
2608 NULL : &timo);
2609 umtxq_remove(uq);
2610 umtxq_unlock(&uq->uq_key);
2611
2612 mtx_lock(&umtx_lock);
2613 uq->uq_inherited_pri = old_inherited_pri;
2614 pri = PRI_MAX;
2615 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
2616 uq2 = TAILQ_FIRST(&pi->pi_blocked);
2617 if (uq2 != NULL) {
2618 if (pri > UPRI(uq2->uq_thread))
2619 pri = UPRI(uq2->uq_thread);
2620 }
2621 }
2622 if (pri > uq->uq_inherited_pri)
2623 pri = uq->uq_inherited_pri;
2624 thread_lock(td);
2625 sched_lend_user_prio(td, pri);
2626 thread_unlock(td);
2627 mtx_unlock(&umtx_lock);
2628 }
2629
2630 if (error != 0 && error != EOWNERDEAD) {
2631 mtx_lock(&umtx_lock);
2632 uq->uq_inherited_pri = old_inherited_pri;
2633 pri = PRI_MAX;
2634 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
2635 uq2 = TAILQ_FIRST(&pi->pi_blocked);
2636 if (uq2 != NULL) {
2637 if (pri > UPRI(uq2->uq_thread))
2638 pri = UPRI(uq2->uq_thread);
2639 }
2640 }
2641 if (pri > uq->uq_inherited_pri)
2642 pri = uq->uq_inherited_pri;
2643 thread_lock(td);
2644 sched_lend_user_prio(td, pri);
2645 thread_unlock(td);
2646 mtx_unlock(&umtx_lock);
2647 }
2648
2649 out:
2650 umtxq_unbusy_unlocked(&uq->uq_key);
2651 umtx_key_release(&uq->uq_key);
2652 return (error);
2653 }
2654
2655 /*
2656 * Unlock a PP mutex.
2657 */
2658 static int
2659 do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags, bool rb)
2660 {
2661 struct umtx_key key;
2662 struct umtx_q *uq, *uq2;
2663 struct umtx_pi *pi;
2664 uint32_t id, owner, rceiling;
2665 int error, pri, new_inherited_pri, su;
2666
2667 id = td->td_tid;
2668 uq = td->td_umtxq;
2669 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0);
2670
2671 /*
2672 * Make sure we own this mtx.
2673 */
2674 error = fueword32(&m->m_owner, &owner);
2675 if (error == -1)
2676 return (EFAULT);
2677
2678 if ((owner & ~UMUTEX_CONTESTED) != id)
2679 return (EPERM);
2680
2681 error = copyin(&m->m_ceilings[1], &rceiling, sizeof(uint32_t));
2682 if (error != 0)
2683 return (error);
2684
2685 if (rceiling == -1)
2686 new_inherited_pri = PRI_MAX;
2687 else {
2688 rceiling = RTP_PRIO_MAX - rceiling;
2689 if (rceiling > RTP_PRIO_MAX)
2690 return (EINVAL);
2691 new_inherited_pri = PRI_MIN_REALTIME + rceiling;
2692 }
2693
2694 if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
2695 TYPE_PP_ROBUST_UMUTEX : TYPE_PP_UMUTEX, GET_SHARE(flags),
2696 &key)) != 0)
2697 return (error);
2698 umtxq_lock(&key);
2699 umtxq_busy(&key);
2700 umtxq_unlock(&key);
2701 /*
2702 * For priority protected mutex, always set unlocked state
2703 * to UMUTEX_CONTESTED, so that userland always enters kernel
2704 * to lock the mutex, it is necessary because thread priority
2705 * has to be adjusted for such mutex.
2706 */
2707 error = suword32(&m->m_owner, umtx_unlock_val(flags, rb) |
2708 UMUTEX_CONTESTED);
2709
2710 umtxq_lock(&key);
2711 if (error == 0)
2712 umtxq_signal(&key, 1);
2713 umtxq_unbusy(&key);
2714 umtxq_unlock(&key);
2715
2716 if (error == -1)
2717 error = EFAULT;
2718 else {
2719 mtx_lock(&umtx_lock);
2720 if (su != 0)
2721 uq->uq_inherited_pri = new_inherited_pri;
2722 pri = PRI_MAX;
2723 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
2724 uq2 = TAILQ_FIRST(&pi->pi_blocked);
2725 if (uq2 != NULL) {
2726 if (pri > UPRI(uq2->uq_thread))
2727 pri = UPRI(uq2->uq_thread);
2728 }
2729 }
2730 if (pri > uq->uq_inherited_pri)
2731 pri = uq->uq_inherited_pri;
2732 thread_lock(td);
2733 sched_lend_user_prio(td, pri);
2734 thread_unlock(td);
2735 mtx_unlock(&umtx_lock);
2736 }
2737 umtx_key_release(&key);
2738 return (error);
2739 }
2740
2741 static int
2742 do_set_ceiling(struct thread *td, struct umutex *m, uint32_t ceiling,
2743 uint32_t *old_ceiling)
2744 {
2745 struct umtx_q *uq;
2746 uint32_t flags, id, owner, save_ceiling;
2747 int error, rv, rv1;
2748
2749 error = fueword32(&m->m_flags, &flags);
2750 if (error == -1)
2751 return (EFAULT);
2752 if ((flags & UMUTEX_PRIO_PROTECT) == 0)
2753 return (EINVAL);
2754 if (ceiling > RTP_PRIO_MAX)
2755 return (EINVAL);
2756 id = td->td_tid;
2757 uq = td->td_umtxq;
2758 if ((error = umtx_key_get(m, (flags & UMUTEX_ROBUST) != 0 ?
2759 TYPE_PP_ROBUST_UMUTEX : TYPE_PP_UMUTEX, GET_SHARE(flags),
2760 &uq->uq_key)) != 0)
2761 return (error);
2762 for (;;) {
2763 umtxq_lock(&uq->uq_key);
2764 umtxq_busy(&uq->uq_key);
2765 umtxq_unlock(&uq->uq_key);
2766
2767 rv = fueword32(&m->m_ceilings[0], &save_ceiling);
2768 if (rv == -1) {
2769 error = EFAULT;
2770 break;
2771 }
2772
2773 rv = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
2774 id | UMUTEX_CONTESTED);
2775 if (rv == -1) {
2776 error = EFAULT;
2777 break;
2778 }
2779
2780 if (rv == 0) {
2781 MPASS(owner == UMUTEX_CONTESTED);
2782 rv = suword32(&m->m_ceilings[0], ceiling);
2783 rv1 = suword32(&m->m_owner, UMUTEX_CONTESTED);
2784 error = (rv == 0 && rv1 == 0) ? 0: EFAULT;
2785 break;
2786 }
2787
2788 if ((owner & ~UMUTEX_CONTESTED) == id) {
2789 rv = suword32(&m->m_ceilings[0], ceiling);
2790 error = rv == 0 ? 0 : EFAULT;
2791 break;
2792 }
2793
2794 if (owner == UMUTEX_RB_OWNERDEAD) {
2795 error = EOWNERDEAD;
2796 break;
2797 } else if (owner == UMUTEX_RB_NOTRECOV) {
2798 error = ENOTRECOVERABLE;
2799 break;
2800 }
2801
2802 /*
2803 * If we caught a signal, we have retried and now
2804 * exit immediately.
2805 */
2806 if (error != 0)
2807 break;
2808
2809 /*
2810 * We set the contested bit, sleep. Otherwise the lock changed
2811 * and we need to retry or we lost a race to the thread
2812 * unlocking the umtx.
2813 */
2814 umtxq_lock(&uq->uq_key);
2815 umtxq_insert(uq);
2816 umtxq_unbusy(&uq->uq_key);
2817 error = umtxq_sleep(uq, "umtxpp", NULL);
2818 umtxq_remove(uq);
2819 umtxq_unlock(&uq->uq_key);
2820 }
2821 umtxq_lock(&uq->uq_key);
2822 if (error == 0)
2823 umtxq_signal(&uq->uq_key, INT_MAX);
2824 umtxq_unbusy(&uq->uq_key);
2825 umtxq_unlock(&uq->uq_key);
2826 umtx_key_release(&uq->uq_key);
2827 if (error == 0 && old_ceiling != NULL) {
2828 rv = suword32(old_ceiling, save_ceiling);
2829 error = rv == 0 ? 0 : EFAULT;
2830 }
2831 return (error);
2832 }
2833
2834 /*
2835 * Lock a userland POSIX mutex.
2836 */
2837 static int
2838 do_lock_umutex(struct thread *td, struct umutex *m,
2839 struct _umtx_time *timeout, int mode)
2840 {
2841 uint32_t flags;
2842 int error;
2843
2844 error = fueword32(&m->m_flags, &flags);
2845 if (error == -1)
2846 return (EFAULT);
2847
2848 switch (flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
2849 case 0:
2850 error = do_lock_normal(td, m, flags, timeout, mode);
2851 break;
2852 case UMUTEX_PRIO_INHERIT:
2853 error = do_lock_pi(td, m, flags, timeout, mode);
2854 break;
2855 case UMUTEX_PRIO_PROTECT:
2856 error = do_lock_pp(td, m, flags, timeout, mode);
2857 break;
2858 default:
2859 return (EINVAL);
2860 }
2861 if (timeout == NULL) {
2862 if (error == EINTR && mode != _UMUTEX_WAIT)
2863 error = ERESTART;
2864 } else {
2865 /* Timed-locking is not restarted. */
2866 if (error == ERESTART)
2867 error = EINTR;
2868 }
2869 return (error);
2870 }
2871
2872 /*
2873 * Unlock a userland POSIX mutex.
2874 */
2875 static int
2876 do_unlock_umutex(struct thread *td, struct umutex *m, bool rb)
2877 {
2878 uint32_t flags;
2879 int error;
2880
2881 error = fueword32(&m->m_flags, &flags);
2882 if (error == -1)
2883 return (EFAULT);
2884
2885 switch (flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
2886 case 0:
2887 return (do_unlock_normal(td, m, flags, rb));
2888 case UMUTEX_PRIO_INHERIT:
2889 return (do_unlock_pi(td, m, flags, rb));
2890 case UMUTEX_PRIO_PROTECT:
2891 return (do_unlock_pp(td, m, flags, rb));
2892 }
2893
2894 return (EINVAL);
2895 }
2896
2897 static int
2898 do_cv_wait(struct thread *td, struct ucond *cv, struct umutex *m,
2899 struct timespec *timeout, u_long wflags)
2900 {
2901 struct abs_timeout timo;
2902 struct umtx_q *uq;
2903 uint32_t flags, clockid, hasw;
2904 int error;
2905
2906 uq = td->td_umtxq;
2907 error = fueword32(&cv->c_flags, &flags);
2908 if (error == -1)
2909 return (EFAULT);
2910 error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &uq->uq_key);
2911 if (error != 0)
2912 return (error);
2913
2914 if ((wflags & CVWAIT_CLOCKID) != 0) {
2915 error = fueword32(&cv->c_clockid, &clockid);
2916 if (error == -1) {
2917 umtx_key_release(&uq->uq_key);
2918 return (EFAULT);
2919 }
2920 if (clockid < CLOCK_REALTIME ||
2921 clockid >= CLOCK_THREAD_CPUTIME_ID) {
2922 /* hmm, only HW clock id will work. */
2923 umtx_key_release(&uq->uq_key);
2924 return (EINVAL);
2925 }
2926 } else {
2927 clockid = CLOCK_REALTIME;
2928 }
2929
2930 umtxq_lock(&uq->uq_key);
2931 umtxq_busy(&uq->uq_key);
2932 umtxq_insert(uq);
2933 umtxq_unlock(&uq->uq_key);
2934
2935 /*
2936 * Set c_has_waiters to 1 before releasing user mutex, also
2937 * don't modify cache line when unnecessary.
2938 */
2939 error = fueword32(&cv->c_has_waiters, &hasw);
2940 if (error == 0 && hasw == 0)
2941 suword32(&cv->c_has_waiters, 1);
2942
2943 umtxq_unbusy_unlocked(&uq->uq_key);
2944
2945 error = do_unlock_umutex(td, m, false);
2946
2947 if (timeout != NULL)
2948 abs_timeout_init(&timo, clockid, (wflags & CVWAIT_ABSTIME) != 0,
2949 timeout);
2950
2951 umtxq_lock(&uq->uq_key);
2952 if (error == 0) {
2953 error = umtxq_sleep(uq, "ucond", timeout == NULL ?
2954 NULL : &timo);
2955 }
2956
2957 if ((uq->uq_flags & UQF_UMTXQ) == 0)
2958 error = 0;
2959 else {
2960 /*
2961 * This must be timeout,interrupted by signal or
2962 * surprious wakeup, clear c_has_waiter flag when
2963 * necessary.
2964 */
2965 umtxq_busy(&uq->uq_key);
2966 if ((uq->uq_flags & UQF_UMTXQ) != 0) {
2967 int oldlen = uq->uq_cur_queue->length;
2968 umtxq_remove(uq);
2969 if (oldlen == 1) {
2970 umtxq_unlock(&uq->uq_key);
2971 suword32(&cv->c_has_waiters, 0);
2972 umtxq_lock(&uq->uq_key);
2973 }
2974 }
2975 umtxq_unbusy(&uq->uq_key);
2976 if (error == ERESTART)
2977 error = EINTR;
2978 }
2979
2980 umtxq_unlock(&uq->uq_key);
2981 umtx_key_release(&uq->uq_key);
2982 return (error);
2983 }
2984
2985 /*
2986 * Signal a userland condition variable.
2987 */
2988 static int
2989 do_cv_signal(struct thread *td, struct ucond *cv)
2990 {
2991 struct umtx_key key;
2992 int error, cnt, nwake;
2993 uint32_t flags;
2994
2995 error = fueword32(&cv->c_flags, &flags);
2996 if (error == -1)
2997 return (EFAULT);
2998 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0)
2999 return (error);
3000 umtxq_lock(&key);
3001 umtxq_busy(&key);
3002 cnt = umtxq_count(&key);
3003 nwake = umtxq_signal(&key, 1);
3004 if (cnt <= nwake) {
3005 umtxq_unlock(&key);
3006 error = suword32(&cv->c_has_waiters, 0);
3007 if (error == -1)
3008 error = EFAULT;
3009 umtxq_lock(&key);
3010 }
3011 umtxq_unbusy(&key);
3012 umtxq_unlock(&key);
3013 umtx_key_release(&key);
3014 return (error);
3015 }
3016
3017 static int
3018 do_cv_broadcast(struct thread *td, struct ucond *cv)
3019 {
3020 struct umtx_key key;
3021 int error;
3022 uint32_t flags;
3023
3024 error = fueword32(&cv->c_flags, &flags);
3025 if (error == -1)
3026 return (EFAULT);
3027 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0)
3028 return (error);
3029
3030 umtxq_lock(&key);
3031 umtxq_busy(&key);
3032 umtxq_signal(&key, INT_MAX);
3033 umtxq_unlock(&key);
3034
3035 error = suword32(&cv->c_has_waiters, 0);
3036 if (error == -1)
3037 error = EFAULT;
3038
3039 umtxq_unbusy_unlocked(&key);
3040
3041 umtx_key_release(&key);
3042 return (error);
3043 }
3044
3045 static int
3046 do_rw_rdlock(struct thread *td, struct urwlock *rwlock, long fflag,
3047 struct _umtx_time *timeout)
3048 {
3049 struct abs_timeout timo;
3050 struct umtx_q *uq;
3051 uint32_t flags, wrflags;
3052 int32_t state, oldstate;
3053 int32_t blocked_readers;
3054 int error, error1, rv;
3055
3056 uq = td->td_umtxq;
3057 error = fueword32(&rwlock->rw_flags, &flags);
3058 if (error == -1)
3059 return (EFAULT);
3060 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
3061 if (error != 0)
3062 return (error);
3063
3064 if (timeout != NULL)
3065 abs_timeout_init2(&timo, timeout);
3066
3067 wrflags = URWLOCK_WRITE_OWNER;
3068 if (!(fflag & URWLOCK_PREFER_READER) && !(flags & URWLOCK_PREFER_READER))
3069 wrflags |= URWLOCK_WRITE_WAITERS;
3070
3071 for (;;) {
3072 rv = fueword32(&rwlock->rw_state, &state);
3073 if (rv == -1) {
3074 umtx_key_release(&uq->uq_key);
3075 return (EFAULT);
3076 }
3077
3078 /* try to lock it */
3079 while (!(state & wrflags)) {
3080 if (__predict_false(URWLOCK_READER_COUNT(state) ==
3081 URWLOCK_MAX_READERS)) {
3082 umtx_key_release(&uq->uq_key);
3083 return (EAGAIN);
3084 }
3085 rv = casueword32(&rwlock->rw_state, state,
3086 &oldstate, state + 1);
3087 if (rv == -1) {
3088 umtx_key_release(&uq->uq_key);
3089 return (EFAULT);
3090 }
3091 if (rv == 0) {
3092 MPASS(oldstate == state);
3093 umtx_key_release(&uq->uq_key);
3094 return (0);
3095 }
3096 error = thread_check_susp(td, true);
3097 if (error != 0)
3098 break;
3099 state = oldstate;
3100 }
3101
3102 if (error)
3103 break;
3104
3105 /* grab monitor lock */
3106 umtxq_lock(&uq->uq_key);
3107 umtxq_busy(&uq->uq_key);
3108 umtxq_unlock(&uq->uq_key);
3109
3110 /*
3111 * re-read the state, in case it changed between the try-lock above
3112 * and the check below
3113 */
3114 rv = fueword32(&rwlock->rw_state, &state);
3115 if (rv == -1)
3116 error = EFAULT;
3117
3118 /* set read contention bit */
3119 while (error == 0 && (state & wrflags) &&
3120 !(state & URWLOCK_READ_WAITERS)) {
3121 rv = casueword32(&rwlock->rw_state, state,
3122 &oldstate, state | URWLOCK_READ_WAITERS);
3123 if (rv == -1) {
3124 error = EFAULT;
3125 break;
3126 }
3127 if (rv == 0) {
3128 MPASS(oldstate == state);
3129 goto sleep;
3130 }
3131 state = oldstate;
3132 error = thread_check_susp(td, false);
3133 if (error != 0)
3134 break;
3135 }
3136 if (error != 0) {
3137 umtxq_unbusy_unlocked(&uq->uq_key);
3138 break;
3139 }
3140
3141 /* state is changed while setting flags, restart */
3142 if (!(state & wrflags)) {
3143 umtxq_unbusy_unlocked(&uq->uq_key);
3144 error = thread_check_susp(td, true);
3145 if (error != 0)
3146 break;
3147 continue;
3148 }
3149
3150 sleep:
3151 /*
3152 * Contention bit is set, before sleeping, increase
3153 * read waiter count.
3154 */
3155 rv = fueword32(&rwlock->rw_blocked_readers,
3156 &blocked_readers);
3157 if (rv == -1) {
3158 umtxq_unbusy_unlocked(&uq->uq_key);
3159 error = EFAULT;
3160 break;
3161 }
3162 suword32(&rwlock->rw_blocked_readers, blocked_readers+1);
3163
3164 while (state & wrflags) {
3165 umtxq_lock(&uq->uq_key);
3166 umtxq_insert(uq);
3167 umtxq_unbusy(&uq->uq_key);
3168
3169 error = umtxq_sleep(uq, "urdlck", timeout == NULL ?
3170 NULL : &timo);
3171
3172 umtxq_busy(&uq->uq_key);
3173 umtxq_remove(uq);
3174 umtxq_unlock(&uq->uq_key);
3175 if (error)
3176 break;
3177 rv = fueword32(&rwlock->rw_state, &state);
3178 if (rv == -1) {
3179 error = EFAULT;
3180 break;
3181 }
3182 }
3183
3184 /* decrease read waiter count, and may clear read contention bit */
3185 rv = fueword32(&rwlock->rw_blocked_readers,
3186 &blocked_readers);
3187 if (rv == -1) {
3188 umtxq_unbusy_unlocked(&uq->uq_key);
3189 error = EFAULT;
3190 break;
3191 }
3192 suword32(&rwlock->rw_blocked_readers, blocked_readers-1);
3193 if (blocked_readers == 1) {
3194 rv = fueword32(&rwlock->rw_state, &state);
3195 if (rv == -1) {
3196 umtxq_unbusy_unlocked(&uq->uq_key);
3197 error = EFAULT;
3198 break;
3199 }
3200 for (;;) {
3201 rv = casueword32(&rwlock->rw_state, state,
3202 &oldstate, state & ~URWLOCK_READ_WAITERS);
3203 if (rv == -1) {
3204 error = EFAULT;
3205 break;
3206 }
3207 if (rv == 0) {
3208 MPASS(oldstate == state);
3209 break;
3210 }
3211 state = oldstate;
3212 error1 = thread_check_susp(td, false);
3213 if (error1 != 0) {
3214 if (error == 0)
3215 error = error1;
3216 break;
3217 }
3218 }
3219 }
3220
3221 umtxq_unbusy_unlocked(&uq->uq_key);
3222 if (error != 0)
3223 break;
3224 }
3225 umtx_key_release(&uq->uq_key);
3226 if (error == ERESTART)
3227 error = EINTR;
3228 return (error);
3229 }
3230
3231 static int
3232 do_rw_wrlock(struct thread *td, struct urwlock *rwlock, struct _umtx_time *timeout)
3233 {
3234 struct abs_timeout timo;
3235 struct umtx_q *uq;
3236 uint32_t flags;
3237 int32_t state, oldstate;
3238 int32_t blocked_writers;
3239 int32_t blocked_readers;
3240 int error, error1, rv;
3241
3242 uq = td->td_umtxq;
3243 error = fueword32(&rwlock->rw_flags, &flags);
3244 if (error == -1)
3245 return (EFAULT);
3246 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
3247 if (error != 0)
3248 return (error);
3249
3250 if (timeout != NULL)
3251 abs_timeout_init2(&timo, timeout);
3252
3253 blocked_readers = 0;
3254 for (;;) {
3255 rv = fueword32(&rwlock->rw_state, &state);
3256 if (rv == -1) {
3257 umtx_key_release(&uq->uq_key);
3258 return (EFAULT);
3259 }
3260 while ((state & URWLOCK_WRITE_OWNER) == 0 &&
3261 URWLOCK_READER_COUNT(state) == 0) {
3262 rv = casueword32(&rwlock->rw_state, state,
3263 &oldstate, state | URWLOCK_WRITE_OWNER);
3264 if (rv == -1) {
3265 umtx_key_release(&uq->uq_key);
3266 return (EFAULT);
3267 }
3268 if (rv == 0) {
3269 MPASS(oldstate == state);
3270 umtx_key_release(&uq->uq_key);
3271 return (0);
3272 }
3273 state = oldstate;
3274 error = thread_check_susp(td, true);
3275 if (error != 0)
3276 break;
3277 }
3278
3279 if (error) {
3280 if ((state & (URWLOCK_WRITE_OWNER |
3281 URWLOCK_WRITE_WAITERS)) == 0 &&
3282 blocked_readers != 0) {
3283 umtxq_lock(&uq->uq_key);
3284 umtxq_busy(&uq->uq_key);
3285 umtxq_signal_queue(&uq->uq_key, INT_MAX,
3286 UMTX_SHARED_QUEUE);
3287 umtxq_unbusy(&uq->uq_key);
3288 umtxq_unlock(&uq->uq_key);
3289 }
3290
3291 break;
3292 }
3293
3294 /* grab monitor lock */
3295 umtxq_lock(&uq->uq_key);
3296 umtxq_busy(&uq->uq_key);
3297 umtxq_unlock(&uq->uq_key);
3298
3299 /*
3300 * Re-read the state, in case it changed between the
3301 * try-lock above and the check below.
3302 */
3303 rv = fueword32(&rwlock->rw_state, &state);
3304 if (rv == -1)
3305 error = EFAULT;
3306
3307 while (error == 0 && ((state & URWLOCK_WRITE_OWNER) ||
3308 URWLOCK_READER_COUNT(state) != 0) &&
3309 (state & URWLOCK_WRITE_WAITERS) == 0) {
3310 rv = casueword32(&rwlock->rw_state, state,
3311 &oldstate, state | URWLOCK_WRITE_WAITERS);
3312 if (rv == -1) {
3313 error = EFAULT;
3314 break;
3315 }
3316 if (rv == 0) {
3317 MPASS(oldstate == state);
3318 goto sleep;
3319 }
3320 state = oldstate;
3321 error = thread_check_susp(td, false);
3322 if (error != 0)
3323 break;
3324 }
3325 if (error != 0) {
3326 umtxq_unbusy_unlocked(&uq->uq_key);
3327 break;
3328 }
3329
3330 if ((state & URWLOCK_WRITE_OWNER) == 0 &&
3331 URWLOCK_READER_COUNT(state) == 0) {
3332 umtxq_unbusy_unlocked(&uq->uq_key);
3333 error = thread_check_susp(td, false);
3334 if (error != 0)
3335 break;
3336 continue;
3337 }
3338 sleep:
3339 rv = fueword32(&rwlock->rw_blocked_writers,
3340 &blocked_writers);
3341 if (rv == -1) {
3342 umtxq_unbusy_unlocked(&uq->uq_key);
3343 error = EFAULT;
3344 break;
3345 }
3346 suword32(&rwlock->rw_blocked_writers, blocked_writers + 1);
3347
3348 while ((state & URWLOCK_WRITE_OWNER) ||
3349 URWLOCK_READER_COUNT(state) != 0) {
3350 umtxq_lock(&uq->uq_key);
3351 umtxq_insert_queue(uq, UMTX_EXCLUSIVE_QUEUE);
3352 umtxq_unbusy(&uq->uq_key);
3353
3354 error = umtxq_sleep(uq, "uwrlck", timeout == NULL ?
3355 NULL : &timo);
3356
3357 umtxq_busy(&uq->uq_key);
3358 umtxq_remove_queue(uq, UMTX_EXCLUSIVE_QUEUE);
3359 umtxq_unlock(&uq->uq_key);
3360 if (error)
3361 break;
3362 rv = fueword32(&rwlock->rw_state, &state);
3363 if (rv == -1) {
3364 error = EFAULT;
3365 break;
3366 }
3367 }
3368
3369 rv = fueword32(&rwlock->rw_blocked_writers,
3370 &blocked_writers);
3371 if (rv == -1) {
3372 umtxq_unbusy_unlocked(&uq->uq_key);
3373 error = EFAULT;
3374 break;
3375 }
3376 suword32(&rwlock->rw_blocked_writers, blocked_writers-1);
3377 if (blocked_writers == 1) {
3378 rv = fueword32(&rwlock->rw_state, &state);
3379 if (rv == -1) {
3380 umtxq_unbusy_unlocked(&uq->uq_key);
3381 error = EFAULT;
3382 break;
3383 }
3384 for (;;) {
3385 rv = casueword32(&rwlock->rw_state, state,
3386 &oldstate, state & ~URWLOCK_WRITE_WAITERS);
3387 if (rv == -1) {
3388 error = EFAULT;
3389 break;
3390 }
3391 if (rv == 0) {
3392 MPASS(oldstate == state);
3393 break;
3394 }
3395 state = oldstate;
3396 error1 = thread_check_susp(td, false);
3397 /*
3398 * We are leaving the URWLOCK_WRITE_WAITERS
3399 * behind, but this should not harm the
3400 * correctness.
3401 */
3402 if (error1 != 0) {
3403 if (error == 0)
3404 error = error1;
3405 break;
3406 }
3407 }
3408 rv = fueword32(&rwlock->rw_blocked_readers,
3409 &blocked_readers);
3410 if (rv == -1) {
3411 umtxq_unbusy_unlocked(&uq->uq_key);
3412 error = EFAULT;
3413 break;
3414 }
3415 } else
3416 blocked_readers = 0;
3417
3418 umtxq_unbusy_unlocked(&uq->uq_key);
3419 }
3420
3421 umtx_key_release(&uq->uq_key);
3422 if (error == ERESTART)
3423 error = EINTR;
3424 return (error);
3425 }
3426
3427 static int
3428 do_rw_unlock(struct thread *td, struct urwlock *rwlock)
3429 {
3430 struct umtx_q *uq;
3431 uint32_t flags;
3432 int32_t state, oldstate;
3433 int error, rv, q, count;
3434
3435 uq = td->td_umtxq;
3436 error = fueword32(&rwlock->rw_flags, &flags);
3437 if (error == -1)
3438 return (EFAULT);
3439 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
3440 if (error != 0)
3441 return (error);
3442
3443 error = fueword32(&rwlock->rw_state, &state);
3444 if (error == -1) {
3445 error = EFAULT;
3446 goto out;
3447 }
3448 if (state & URWLOCK_WRITE_OWNER) {
3449 for (;;) {
3450 rv = casueword32(&rwlock->rw_state, state,
3451 &oldstate, state & ~URWLOCK_WRITE_OWNER);
3452 if (rv == -1) {
3453 error = EFAULT;
3454 goto out;
3455 }
3456 if (rv == 1) {
3457 state = oldstate;
3458 if (!(oldstate & URWLOCK_WRITE_OWNER)) {
3459 error = EPERM;
3460 goto out;
3461 }
3462 error = thread_check_susp(td, true);
3463 if (error != 0)
3464 goto out;
3465 } else
3466 break;
3467 }
3468 } else if (URWLOCK_READER_COUNT(state) != 0) {
3469 for (;;) {
3470 rv = casueword32(&rwlock->rw_state, state,
3471 &oldstate, state - 1);
3472 if (rv == -1) {
3473 error = EFAULT;
3474 goto out;
3475 }
3476 if (rv == 1) {
3477 state = oldstate;
3478 if (URWLOCK_READER_COUNT(oldstate) == 0) {
3479 error = EPERM;
3480 goto out;
3481 }
3482 error = thread_check_susp(td, true);
3483 if (error != 0)
3484 goto out;
3485 } else
3486 break;
3487 }
3488 } else {
3489 error = EPERM;
3490 goto out;
3491 }
3492
3493 count = 0;
3494
3495 if (!(flags & URWLOCK_PREFER_READER)) {
3496 if (state & URWLOCK_WRITE_WAITERS) {
3497 count = 1;
3498 q = UMTX_EXCLUSIVE_QUEUE;
3499 } else if (state & URWLOCK_READ_WAITERS) {
3500 count = INT_MAX;
3501 q = UMTX_SHARED_QUEUE;
3502 }
3503 } else {
3504 if (state & URWLOCK_READ_WAITERS) {
3505 count = INT_MAX;
3506 q = UMTX_SHARED_QUEUE;
3507 } else if (state & URWLOCK_WRITE_WAITERS) {
3508 count = 1;
3509 q = UMTX_EXCLUSIVE_QUEUE;
3510 }
3511 }
3512
3513 if (count) {
3514 umtxq_lock(&uq->uq_key);
3515 umtxq_busy(&uq->uq_key);
3516 umtxq_signal_queue(&uq->uq_key, count, q);
3517 umtxq_unbusy(&uq->uq_key);
3518 umtxq_unlock(&uq->uq_key);
3519 }
3520 out:
3521 umtx_key_release(&uq->uq_key);
3522 return (error);
3523 }
3524
3525 #if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
3526 static int
3527 do_sem_wait(struct thread *td, struct _usem *sem, struct _umtx_time *timeout)
3528 {
3529 struct abs_timeout timo;
3530 struct umtx_q *uq;
3531 uint32_t flags, count, count1;
3532 int error, rv, rv1;
3533
3534 uq = td->td_umtxq;
3535 error = fueword32(&sem->_flags, &flags);
3536 if (error == -1)
3537 return (EFAULT);
3538 error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key);
3539 if (error != 0)
3540 return (error);
3541
3542 if (timeout != NULL)
3543 abs_timeout_init2(&timo, timeout);
3544
3545 again:
3546 umtxq_lock(&uq->uq_key);
3547 umtxq_busy(&uq->uq_key);
3548 umtxq_insert(uq);
3549 umtxq_unlock(&uq->uq_key);
3550 rv = casueword32(&sem->_has_waiters, 0, &count1, 1);
3551 if (rv == 0)
3552 rv1 = fueword32(&sem->_count, &count);
3553 if (rv == -1 || (rv == 0 && (rv1 == -1 || count != 0)) ||
3554 (rv == 1 && count1 == 0)) {
3555 umtxq_lock(&uq->uq_key);
3556 umtxq_unbusy(&uq->uq_key);
3557 umtxq_remove(uq);
3558 umtxq_unlock(&uq->uq_key);
3559 if (rv == 1) {
3560 rv = thread_check_susp(td, true);
3561 if (rv == 0)
3562 goto again;
3563 error = rv;
3564 goto out;
3565 }
3566 if (rv == 0)
3567 rv = rv1;
3568 error = rv == -1 ? EFAULT : 0;
3569 goto out;
3570 }
3571 umtxq_lock(&uq->uq_key);
3572 umtxq_unbusy(&uq->uq_key);
3573
3574 error = umtxq_sleep(uq, "usem", timeout == NULL ? NULL : &timo);
3575
3576 if ((uq->uq_flags & UQF_UMTXQ) == 0)
3577 error = 0;
3578 else {
3579 umtxq_remove(uq);
3580 /* A relative timeout cannot be restarted. */
3581 if (error == ERESTART && timeout != NULL &&
3582 (timeout->_flags & UMTX_ABSTIME) == 0)
3583 error = EINTR;
3584 }
3585 umtxq_unlock(&uq->uq_key);
3586 out:
3587 umtx_key_release(&uq->uq_key);
3588 return (error);
3589 }
3590
3591 /*
3592 * Signal a userland semaphore.
3593 */
3594 static int
3595 do_sem_wake(struct thread *td, struct _usem *sem)
3596 {
3597 struct umtx_key key;
3598 int error, cnt;
3599 uint32_t flags;
3600
3601 error = fueword32(&sem->_flags, &flags);
3602 if (error == -1)
3603 return (EFAULT);
3604 if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0)
3605 return (error);
3606 umtxq_lock(&key);
3607 umtxq_busy(&key);
3608 cnt = umtxq_count(&key);
3609 if (cnt > 0) {
3610 /*
3611 * Check if count is greater than 0, this means the memory is
3612 * still being referenced by user code, so we can safely
3613 * update _has_waiters flag.
3614 */
3615 if (cnt == 1) {
3616 umtxq_unlock(&key);
3617 error = suword32(&sem->_has_waiters, 0);
3618 umtxq_lock(&key);
3619 if (error == -1)
3620 error = EFAULT;
3621 }
3622 umtxq_signal(&key, 1);
3623 }
3624 umtxq_unbusy(&key);
3625 umtxq_unlock(&key);
3626 umtx_key_release(&key);
3627 return (error);
3628 }
3629 #endif
3630
3631 static int
3632 do_sem2_wait(struct thread *td, struct _usem2 *sem, struct _umtx_time *timeout)
3633 {
3634 struct abs_timeout timo;
3635 struct umtx_q *uq;
3636 uint32_t count, flags;
3637 int error, rv;
3638
3639 uq = td->td_umtxq;
3640 flags = fuword32(&sem->_flags);
3641 if (timeout != NULL)
3642 abs_timeout_init2(&timo, timeout);
3643
3644 again:
3645 error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key);
3646 if (error != 0)
3647 return (error);
3648 umtxq_lock(&uq->uq_key);
3649 umtxq_busy(&uq->uq_key);
3650 umtxq_insert(uq);
3651 umtxq_unlock(&uq->uq_key);
3652 rv = fueword32(&sem->_count, &count);
3653 if (rv == -1) {
3654 umtxq_lock(&uq->uq_key);
3655 umtxq_unbusy(&uq->uq_key);
3656 umtxq_remove(uq);
3657 umtxq_unlock(&uq->uq_key);
3658 umtx_key_release(&uq->uq_key);
3659 return (EFAULT);
3660 }
3661 for (;;) {
3662 if (USEM_COUNT(count) != 0) {
3663 umtxq_lock(&uq->uq_key);
3664 umtxq_unbusy(&uq->uq_key);
3665 umtxq_remove(uq);
3666 umtxq_unlock(&uq->uq_key);
3667 umtx_key_release(&uq->uq_key);
3668 return (0);
3669 }
3670 if (count == USEM_HAS_WAITERS)
3671 break;
3672 rv = casueword32(&sem->_count, 0, &count, USEM_HAS_WAITERS);
3673 if (rv == 0)
3674 break;
3675 umtxq_lock(&uq->uq_key);
3676 umtxq_unbusy(&uq->uq_key);
3677 umtxq_remove(uq);
3678 umtxq_unlock(&uq->uq_key);
3679 umtx_key_release(&uq->uq_key);
3680 if (rv == -1)
3681 return (EFAULT);
3682 rv = thread_check_susp(td, true);
3683 if (rv != 0)
3684 return (rv);
3685 goto again;
3686 }
3687 umtxq_lock(&uq->uq_key);
3688 umtxq_unbusy(&uq->uq_key);
3689
3690 error = umtxq_sleep(uq, "usem", timeout == NULL ? NULL : &timo);
3691
3692 if ((uq->uq_flags & UQF_UMTXQ) == 0)
3693 error = 0;
3694 else {
3695 umtxq_remove(uq);
3696 if (timeout != NULL && (timeout->_flags & UMTX_ABSTIME) == 0) {
3697 /* A relative timeout cannot be restarted. */
3698 if (error == ERESTART)
3699 error = EINTR;
3700 if (error == EINTR) {
3701 abs_timeout_update(&timo);
3702 timespecsub(&timo.end, &timo.cur,
3703 &timeout->_timeout);
3704 }
3705 }
3706 }
3707 umtxq_unlock(&uq->uq_key);
3708 umtx_key_release(&uq->uq_key);
3709 return (error);
3710 }
3711
3712 /*
3713 * Signal a userland semaphore.
3714 */
3715 static int
3716 do_sem2_wake(struct thread *td, struct _usem2 *sem)
3717 {
3718 struct umtx_key key;
3719 int error, cnt, rv;
3720 uint32_t count, flags;
3721
3722 rv = fueword32(&sem->_flags, &flags);
3723 if (rv == -1)
3724 return (EFAULT);
3725 if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0)
3726 return (error);
3727 umtxq_lock(&key);
3728 umtxq_busy(&key);
3729 cnt = umtxq_count(&key);
3730 if (cnt > 0) {
3731 /*
3732 * If this was the last sleeping thread, clear the waiters
3733 * flag in _count.
3734 */
3735 if (cnt == 1) {
3736 umtxq_unlock(&key);
3737 rv = fueword32(&sem->_count, &count);
3738 while (rv != -1 && count & USEM_HAS_WAITERS) {
3739 rv = casueword32(&sem->_count, count, &count,
3740 count & ~USEM_HAS_WAITERS);
3741 if (rv == 1) {
3742 rv = thread_check_susp(td, true);
3743 if (rv != 0)
3744 break;
3745 }
3746 }
3747 if (rv == -1)
3748 error = EFAULT;
3749 else if (rv > 0) {
3750 error = rv;
3751 }
3752 umtxq_lock(&key);
3753 }
3754
3755 umtxq_signal(&key, 1);
3756 }
3757 umtxq_unbusy(&key);
3758 umtxq_unlock(&key);
3759 umtx_key_release(&key);
3760 return (error);
3761 }
3762
3763 #ifdef COMPAT_FREEBSD10
3764 int
3765 freebsd10__umtx_lock(struct thread *td, struct freebsd10__umtx_lock_args *uap)
3766 {
3767 return (do_lock_umtx(td, uap->umtx, td->td_tid, 0));
3768 }
3769
3770 int
3771 freebsd10__umtx_unlock(struct thread *td,
3772 struct freebsd10__umtx_unlock_args *uap)
3773 {
3774 return (do_unlock_umtx(td, uap->umtx, td->td_tid));
3775 }
3776 #endif
3777
3778 inline int
3779 umtx_copyin_timeout(const void *uaddr, struct timespec *tsp)
3780 {
3781 int error;
3782
3783 error = copyin(uaddr, tsp, sizeof(*tsp));
3784 if (error == 0) {
3785 if (tsp->tv_sec < 0 ||
3786 tsp->tv_nsec >= 1000000000 ||
3787 tsp->tv_nsec < 0)
3788 error = EINVAL;
3789 }
3790 return (error);
3791 }
3792
3793 static inline int
3794 umtx_copyin_umtx_time(const void *uaddr, size_t size, struct _umtx_time *tp)
3795 {
3796 int error;
3797
3798 if (size <= sizeof(tp->_timeout)) {
3799 tp->_clockid = CLOCK_REALTIME;
3800 tp->_flags = 0;
3801 error = copyin(uaddr, &tp->_timeout, sizeof(tp->_timeout));
3802 } else
3803 error = copyin(uaddr, tp, sizeof(*tp));
3804 if (error != 0)
3805 return (error);
3806 if (tp->_timeout.tv_sec < 0 ||
3807 tp->_timeout.tv_nsec >= 1000000000 || tp->_timeout.tv_nsec < 0)
3808 return (EINVAL);
3809 return (0);
3810 }
3811
3812 static int
3813 umtx_copyin_robust_lists(const void *uaddr, size_t size,
3814 struct umtx_robust_lists_params *rb)
3815 {
3816
3817 if (size > sizeof(*rb))
3818 return (EINVAL);
3819 return (copyin(uaddr, rb, size));
3820 }
3821
3822 static int
3823 umtx_copyout_timeout(void *uaddr, size_t sz, struct timespec *tsp)
3824 {
3825
3826 /*
3827 * Should be guaranteed by the caller, sz == uaddr1 - sizeof(_umtx_time)
3828 * and we're only called if sz >= sizeof(timespec) as supplied in the
3829 * copyops.
3830 */
3831 KASSERT(sz >= sizeof(*tsp),
3832 ("umtx_copyops specifies incorrect sizes"));
3833
3834 return (copyout(tsp, uaddr, sizeof(*tsp)));
3835 }
3836
3837 #ifdef COMPAT_FREEBSD10
3838 static int
3839 __umtx_op_lock_umtx(struct thread *td, struct _umtx_op_args *uap,
3840 const struct umtx_copyops *ops)
3841 {
3842 struct timespec *ts, timeout;
3843 int error;
3844
3845 /* Allow a null timespec (wait forever). */
3846 if (uap->uaddr2 == NULL)
3847 ts = NULL;
3848 else {
3849 error = ops->copyin_timeout(uap->uaddr2, &timeout);
3850 if (error != 0)
3851 return (error);
3852 ts = &timeout;
3853 }
3854 #ifdef COMPAT_FREEBSD32
3855 if (ops->compat32)
3856 return (do_lock_umtx32(td, uap->obj, uap->val, ts));
3857 #endif
3858 return (do_lock_umtx(td, uap->obj, uap->val, ts));
3859 }
3860
3861 static int
3862 __umtx_op_unlock_umtx(struct thread *td, struct _umtx_op_args *uap,
3863 const struct umtx_copyops *ops)
3864 {
3865 #ifdef COMPAT_FREEBSD32
3866 if (ops->compat32)
3867 return (do_unlock_umtx32(td, uap->obj, uap->val));
3868 #endif
3869 return (do_unlock_umtx(td, uap->obj, uap->val));
3870 }
3871 #endif /* COMPAT_FREEBSD10 */
3872
3873 #if !defined(COMPAT_FREEBSD10)
3874 static int
3875 __umtx_op_unimpl(struct thread *td __unused, struct _umtx_op_args *uap __unused,
3876 const struct umtx_copyops *ops __unused)
3877 {
3878 return (EOPNOTSUPP);
3879 }
3880 #endif /* COMPAT_FREEBSD10 */
3881
3882 static int
3883 __umtx_op_wait(struct thread *td, struct _umtx_op_args *uap,
3884 const struct umtx_copyops *ops)
3885 {
3886 struct _umtx_time timeout, *tm_p;
3887 int error;
3888
3889 if (uap->uaddr2 == NULL)
3890 tm_p = NULL;
3891 else {
3892 error = ops->copyin_umtx_time(
3893 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3894 if (error != 0)
3895 return (error);
3896 tm_p = &timeout;
3897 }
3898 return (do_wait(td, uap->obj, uap->val, tm_p, ops->compat32, 0));
3899 }
3900
3901 static int
3902 __umtx_op_wait_uint(struct thread *td, struct _umtx_op_args *uap,
3903 const struct umtx_copyops *ops)
3904 {
3905 struct _umtx_time timeout, *tm_p;
3906 int error;
3907
3908 if (uap->uaddr2 == NULL)
3909 tm_p = NULL;
3910 else {
3911 error = ops->copyin_umtx_time(
3912 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3913 if (error != 0)
3914 return (error);
3915 tm_p = &timeout;
3916 }
3917 return (do_wait(td, uap->obj, uap->val, tm_p, 1, 0));
3918 }
3919
3920 static int
3921 __umtx_op_wait_uint_private(struct thread *td, struct _umtx_op_args *uap,
3922 const struct umtx_copyops *ops)
3923 {
3924 struct _umtx_time *tm_p, timeout;
3925 int error;
3926
3927 if (uap->uaddr2 == NULL)
3928 tm_p = NULL;
3929 else {
3930 error = ops->copyin_umtx_time(
3931 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3932 if (error != 0)
3933 return (error);
3934 tm_p = &timeout;
3935 }
3936 return (do_wait(td, uap->obj, uap->val, tm_p, 1, 1));
3937 }
3938
3939 static int
3940 __umtx_op_wake(struct thread *td, struct _umtx_op_args *uap,
3941 const struct umtx_copyops *ops __unused)
3942 {
3943
3944 return (kern_umtx_wake(td, uap->obj, uap->val, 0));
3945 }
3946
3947 #define BATCH_SIZE 128
3948 static int
3949 __umtx_op_nwake_private_native(struct thread *td, struct _umtx_op_args *uap)
3950 {
3951 char *uaddrs[BATCH_SIZE], **upp;
3952 int count, error, i, pos, tocopy;
3953
3954 upp = (char **)uap->obj;
3955 error = 0;
3956 for (count = uap->val, pos = 0; count > 0; count -= tocopy,
3957 pos += tocopy) {
3958 tocopy = MIN(count, BATCH_SIZE);
3959 error = copyin(upp + pos, uaddrs, tocopy * sizeof(char *));
3960 if (error != 0)
3961 break;
3962 for (i = 0; i < tocopy; ++i) {
3963 kern_umtx_wake(td, uaddrs[i], INT_MAX, 1);
3964 }
3965 maybe_yield();
3966 }
3967 return (error);
3968 }
3969
3970 static int
3971 __umtx_op_nwake_private_compat32(struct thread *td, struct _umtx_op_args *uap)
3972 {
3973 uint32_t uaddrs[BATCH_SIZE], *upp;
3974 int count, error, i, pos, tocopy;
3975
3976 upp = (uint32_t *)uap->obj;
3977 error = 0;
3978 for (count = uap->val, pos = 0; count > 0; count -= tocopy,
3979 pos += tocopy) {
3980 tocopy = MIN(count, BATCH_SIZE);
3981 error = copyin(upp + pos, uaddrs, tocopy * sizeof(uint32_t));
3982 if (error != 0)
3983 break;
3984 for (i = 0; i < tocopy; ++i) {
3985 kern_umtx_wake(td, (void *)(uintptr_t)uaddrs[i],
3986 INT_MAX, 1);
3987 }
3988 maybe_yield();
3989 }
3990 return (error);
3991 }
3992
3993 static int
3994 __umtx_op_nwake_private(struct thread *td, struct _umtx_op_args *uap,
3995 const struct umtx_copyops *ops)
3996 {
3997
3998 if (ops->compat32)
3999 return (__umtx_op_nwake_private_compat32(td, uap));
4000 return (__umtx_op_nwake_private_native(td, uap));
4001 }
4002
4003 static int
4004 __umtx_op_wake_private(struct thread *td, struct _umtx_op_args *uap,
4005 const struct umtx_copyops *ops __unused)
4006 {
4007
4008 return (kern_umtx_wake(td, uap->obj, uap->val, 1));
4009 }
4010
4011 static int
4012 __umtx_op_lock_umutex(struct thread *td, struct _umtx_op_args *uap,
4013 const struct umtx_copyops *ops)
4014 {
4015 struct _umtx_time *tm_p, timeout;
4016 int error;
4017
4018 /* Allow a null timespec (wait forever). */
4019 if (uap->uaddr2 == NULL)
4020 tm_p = NULL;
4021 else {
4022 error = ops->copyin_umtx_time(
4023 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
4024 if (error != 0)
4025 return (error);
4026 tm_p = &timeout;
4027 }
4028 return (do_lock_umutex(td, uap->obj, tm_p, 0));
4029 }
4030
4031 static int
4032 __umtx_op_trylock_umutex(struct thread *td, struct _umtx_op_args *uap,
4033 const struct umtx_copyops *ops __unused)
4034 {
4035
4036 return (do_lock_umutex(td, uap->obj, NULL, _UMUTEX_TRY));
4037 }
4038
4039 static int
4040 __umtx_op_wait_umutex(struct thread *td, struct _umtx_op_args *uap,
4041 const struct umtx_copyops *ops)
4042 {
4043 struct _umtx_time *tm_p, timeout;
4044 int error;
4045
4046 /* Allow a null timespec (wait forever). */
4047 if (uap->uaddr2 == NULL)
4048 tm_p = NULL;
4049 else {
4050 error = ops->copyin_umtx_time(
4051 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
4052 if (error != 0)
4053 return (error);
4054 tm_p = &timeout;
4055 }
4056 return (do_lock_umutex(td, uap->obj, tm_p, _UMUTEX_WAIT));
4057 }
4058
4059 static int
4060 __umtx_op_wake_umutex(struct thread *td, struct _umtx_op_args *uap,
4061 const struct umtx_copyops *ops __unused)
4062 {
4063
4064 return (do_wake_umutex(td, uap->obj));
4065 }
4066
4067 static int
4068 __umtx_op_unlock_umutex(struct thread *td, struct _umtx_op_args *uap,
4069 const struct umtx_copyops *ops __unused)
4070 {
4071
4072 return (do_unlock_umutex(td, uap->obj, false));
4073 }
4074
4075 static int
4076 __umtx_op_set_ceiling(struct thread *td, struct _umtx_op_args *uap,
4077 const struct umtx_copyops *ops __unused)
4078 {
4079
4080 return (do_set_ceiling(td, uap->obj, uap->val, uap->uaddr1));
4081 }
4082
4083 static int
4084 __umtx_op_cv_wait(struct thread *td, struct _umtx_op_args *uap,
4085 const struct umtx_copyops *ops)
4086 {
4087 struct timespec *ts, timeout;
4088 int error;
4089
4090 /* Allow a null timespec (wait forever). */
4091 if (uap->uaddr2 == NULL)
4092 ts = NULL;
4093 else {
4094 error = ops->copyin_timeout(uap->uaddr2, &timeout);
4095 if (error != 0)
4096 return (error);
4097 ts = &timeout;
4098 }
4099 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val));
4100 }
4101
4102 static int
4103 __umtx_op_cv_signal(struct thread *td, struct _umtx_op_args *uap,
4104 const struct umtx_copyops *ops __unused)
4105 {
4106
4107 return (do_cv_signal(td, uap->obj));
4108 }
4109
4110 static int
4111 __umtx_op_cv_broadcast(struct thread *td, struct _umtx_op_args *uap,
4112 const struct umtx_copyops *ops __unused)
4113 {
4114
4115 return (do_cv_broadcast(td, uap->obj));
4116 }
4117
4118 static int
4119 __umtx_op_rw_rdlock(struct thread *td, struct _umtx_op_args *uap,
4120 const struct umtx_copyops *ops)
4121 {
4122 struct _umtx_time timeout;
4123 int error;
4124
4125 /* Allow a null timespec (wait forever). */
4126 if (uap->uaddr2 == NULL) {
4127 error = do_rw_rdlock(td, uap->obj, uap->val, 0);
4128 } else {
4129 error = ops->copyin_umtx_time(uap->uaddr2,
4130 (size_t)uap->uaddr1, &timeout);
4131 if (error != 0)
4132 return (error);
4133 error = do_rw_rdlock(td, uap->obj, uap->val, &timeout);
4134 }
4135 return (error);
4136 }
4137
4138 static int
4139 __umtx_op_rw_wrlock(struct thread *td, struct _umtx_op_args *uap,
4140 const struct umtx_copyops *ops)
4141 {
4142 struct _umtx_time timeout;
4143 int error;
4144
4145 /* Allow a null timespec (wait forever). */
4146 if (uap->uaddr2 == NULL) {
4147 error = do_rw_wrlock(td, uap->obj, 0);
4148 } else {
4149 error = ops->copyin_umtx_time(uap->uaddr2,
4150 (size_t)uap->uaddr1, &timeout);
4151 if (error != 0)
4152 return (error);
4153
4154 error = do_rw_wrlock(td, uap->obj, &timeout);
4155 }
4156 return (error);
4157 }
4158
4159 static int
4160 __umtx_op_rw_unlock(struct thread *td, struct _umtx_op_args *uap,
4161 const struct umtx_copyops *ops __unused)
4162 {
4163
4164 return (do_rw_unlock(td, uap->obj));
4165 }
4166
4167 #if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
4168 static int
4169 __umtx_op_sem_wait(struct thread *td, struct _umtx_op_args *uap,
4170 const struct umtx_copyops *ops)
4171 {
4172 struct _umtx_time *tm_p, timeout;
4173 int error;
4174
4175 /* Allow a null timespec (wait forever). */
4176 if (uap->uaddr2 == NULL)
4177 tm_p = NULL;
4178 else {
4179 error = ops->copyin_umtx_time(
4180 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
4181 if (error != 0)
4182 return (error);
4183 tm_p = &timeout;
4184 }
4185 return (do_sem_wait(td, uap->obj, tm_p));
4186 }
4187
4188 static int
4189 __umtx_op_sem_wake(struct thread *td, struct _umtx_op_args *uap,
4190 const struct umtx_copyops *ops __unused)
4191 {
4192
4193 return (do_sem_wake(td, uap->obj));
4194 }
4195 #endif
4196
4197 static int
4198 __umtx_op_wake2_umutex(struct thread *td, struct _umtx_op_args *uap,
4199 const struct umtx_copyops *ops __unused)
4200 {
4201
4202 return (do_wake2_umutex(td, uap->obj, uap->val));
4203 }
4204
4205 static int
4206 __umtx_op_sem2_wait(struct thread *td, struct _umtx_op_args *uap,
4207 const struct umtx_copyops *ops)
4208 {
4209 struct _umtx_time *tm_p, timeout;
4210 size_t uasize;
4211 int error;
4212
4213 /* Allow a null timespec (wait forever). */
4214 if (uap->uaddr2 == NULL) {
4215 uasize = 0;
4216 tm_p = NULL;
4217 } else {
4218 uasize = (size_t)uap->uaddr1;
4219 error = ops->copyin_umtx_time(uap->uaddr2, uasize, &timeout);
4220 if (error != 0)
4221 return (error);
4222 tm_p = &timeout;
4223 }
4224 error = do_sem2_wait(td, uap->obj, tm_p);
4225 if (error == EINTR && uap->uaddr2 != NULL &&
4226 (timeout._flags & UMTX_ABSTIME) == 0 &&
4227 uasize >= ops->umtx_time_sz + ops->timespec_sz) {
4228 error = ops->copyout_timeout(
4229 (void *)((uintptr_t)uap->uaddr2 + ops->umtx_time_sz),
4230 uasize - ops->umtx_time_sz, &timeout._timeout);
4231 if (error == 0) {
4232 error = EINTR;
4233 }
4234 }
4235
4236 return (error);
4237 }
4238
4239 static int
4240 __umtx_op_sem2_wake(struct thread *td, struct _umtx_op_args *uap,
4241 const struct umtx_copyops *ops __unused)
4242 {
4243
4244 return (do_sem2_wake(td, uap->obj));
4245 }
4246
4247 #define USHM_OBJ_UMTX(o) \
4248 ((struct umtx_shm_obj_list *)(&(o)->umtx_data))
4249
4250 #define USHMF_REG_LINKED 0x0001
4251 #define USHMF_OBJ_LINKED 0x0002
4252 struct umtx_shm_reg {
4253 TAILQ_ENTRY(umtx_shm_reg) ushm_reg_link;
4254 LIST_ENTRY(umtx_shm_reg) ushm_obj_link;
4255 struct umtx_key ushm_key;
4256 struct ucred *ushm_cred;
4257 struct shmfd *ushm_obj;
4258 u_int ushm_refcnt;
4259 u_int ushm_flags;
4260 };
4261
4262 LIST_HEAD(umtx_shm_obj_list, umtx_shm_reg);
4263 TAILQ_HEAD(umtx_shm_reg_head, umtx_shm_reg);
4264
4265 static uma_zone_t umtx_shm_reg_zone;
4266 static struct umtx_shm_reg_head umtx_shm_registry[UMTX_CHAINS];
4267 static struct mtx umtx_shm_lock;
4268 static struct umtx_shm_reg_head umtx_shm_reg_delfree =
4269 TAILQ_HEAD_INITIALIZER(umtx_shm_reg_delfree);
4270
4271 static void umtx_shm_free_reg(struct umtx_shm_reg *reg);
4272
4273 static void
4274 umtx_shm_reg_delfree_tq(void *context __unused, int pending __unused)
4275 {
4276 struct umtx_shm_reg_head d;
4277 struct umtx_shm_reg *reg, *reg1;
4278
4279 TAILQ_INIT(&d);
4280 mtx_lock(&umtx_shm_lock);
4281 TAILQ_CONCAT(&d, &umtx_shm_reg_delfree, ushm_reg_link);
4282 mtx_unlock(&umtx_shm_lock);
4283 TAILQ_FOREACH_SAFE(reg, &d, ushm_reg_link, reg1) {
4284 TAILQ_REMOVE(&d, reg, ushm_reg_link);
4285 umtx_shm_free_reg(reg);
4286 }
4287 }
4288
4289 static struct task umtx_shm_reg_delfree_task =
4290 TASK_INITIALIZER(0, umtx_shm_reg_delfree_tq, NULL);
4291
4292 static struct umtx_shm_reg *
4293 umtx_shm_find_reg_locked(const struct umtx_key *key)
4294 {
4295 struct umtx_shm_reg *reg;
4296 struct umtx_shm_reg_head *reg_head;
4297
4298 KASSERT(key->shared, ("umtx_p_find_rg: private key"));
4299 mtx_assert(&umtx_shm_lock, MA_OWNED);
4300 reg_head = &umtx_shm_registry[key->hash];
4301 TAILQ_FOREACH(reg, reg_head, ushm_reg_link) {
4302 KASSERT(reg->ushm_key.shared,
4303 ("non-shared key on reg %p %d", reg, reg->ushm_key.shared));
4304 if (reg->ushm_key.info.shared.object ==
4305 key->info.shared.object &&
4306 reg->ushm_key.info.shared.offset ==
4307 key->info.shared.offset) {
4308 KASSERT(reg->ushm_key.type == TYPE_SHM, ("TYPE_USHM"));
4309 KASSERT(reg->ushm_refcnt > 0,
4310 ("reg %p refcnt 0 onlist", reg));
4311 KASSERT((reg->ushm_flags & USHMF_REG_LINKED) != 0,
4312 ("reg %p not linked", reg));
4313 reg->ushm_refcnt++;
4314 return (reg);
4315 }
4316 }
4317 return (NULL);
4318 }
4319
4320 static struct umtx_shm_reg *
4321 umtx_shm_find_reg(const struct umtx_key *key)
4322 {
4323 struct umtx_shm_reg *reg;
4324
4325 mtx_lock(&umtx_shm_lock);
4326 reg = umtx_shm_find_reg_locked(key);
4327 mtx_unlock(&umtx_shm_lock);
4328 return (reg);
4329 }
4330
4331 static void
4332 umtx_shm_free_reg(struct umtx_shm_reg *reg)
4333 {
4334
4335 chgumtxcnt(reg->ushm_cred->cr_ruidinfo, -1, 0);
4336 crfree(reg->ushm_cred);
4337 shm_drop(reg->ushm_obj);
4338 uma_zfree(umtx_shm_reg_zone, reg);
4339 }
4340
4341 static bool
4342 umtx_shm_unref_reg_locked(struct umtx_shm_reg *reg, bool force)
4343 {
4344 bool res;
4345
4346 mtx_assert(&umtx_shm_lock, MA_OWNED);
4347 KASSERT(reg->ushm_refcnt > 0, ("ushm_reg %p refcnt 0", reg));
4348 reg->ushm_refcnt--;
4349 res = reg->ushm_refcnt == 0;
4350 if (res || force) {
4351 if ((reg->ushm_flags & USHMF_REG_LINKED) != 0) {
4352 TAILQ_REMOVE(&umtx_shm_registry[reg->ushm_key.hash],
4353 reg, ushm_reg_link);
4354 reg->ushm_flags &= ~USHMF_REG_LINKED;
4355 }
4356 if ((reg->ushm_flags & USHMF_OBJ_LINKED) != 0) {
4357 LIST_REMOVE(reg, ushm_obj_link);
4358 reg->ushm_flags &= ~USHMF_OBJ_LINKED;
4359 }
4360 }
4361 return (res);
4362 }
4363
4364 static void
4365 umtx_shm_unref_reg(struct umtx_shm_reg *reg, bool force)
4366 {
4367 vm_object_t object;
4368 bool dofree;
4369
4370 if (force) {
4371 object = reg->ushm_obj->shm_object;
4372 VM_OBJECT_WLOCK(object);
4373 object->flags |= OBJ_UMTXDEAD;
4374 VM_OBJECT_WUNLOCK(object);
4375 }
4376 mtx_lock(&umtx_shm_lock);
4377 dofree = umtx_shm_unref_reg_locked(reg, force);
4378 mtx_unlock(&umtx_shm_lock);
4379 if (dofree)
4380 umtx_shm_free_reg(reg);
4381 }
4382
4383 void
4384 umtx_shm_object_init(vm_object_t object)
4385 {
4386
4387 LIST_INIT(USHM_OBJ_UMTX(object));
4388 }
4389
4390 void
4391 umtx_shm_object_terminated(vm_object_t object)
4392 {
4393 struct umtx_shm_reg *reg, *reg1;
4394 bool dofree;
4395
4396 if (LIST_EMPTY(USHM_OBJ_UMTX(object)))
4397 return;
4398
4399 dofree = false;
4400 mtx_lock(&umtx_shm_lock);
4401 LIST_FOREACH_SAFE(reg, USHM_OBJ_UMTX(object), ushm_obj_link, reg1) {
4402 if (umtx_shm_unref_reg_locked(reg, true)) {
4403 TAILQ_INSERT_TAIL(&umtx_shm_reg_delfree, reg,
4404 ushm_reg_link);
4405 dofree = true;
4406 }
4407 }
4408 mtx_unlock(&umtx_shm_lock);
4409 if (dofree)
4410 taskqueue_enqueue(taskqueue_thread, &umtx_shm_reg_delfree_task);
4411 }
4412
4413 static int
4414 umtx_shm_create_reg(struct thread *td, const struct umtx_key *key,
4415 struct umtx_shm_reg **res)
4416 {
4417 struct umtx_shm_reg *reg, *reg1;
4418 struct ucred *cred;
4419 int error;
4420
4421 reg = umtx_shm_find_reg(key);
4422 if (reg != NULL) {
4423 *res = reg;
4424 return (0);
4425 }
4426 cred = td->td_ucred;
4427 if (!chgumtxcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_UMTXP)))
4428 return (ENOMEM);
4429 reg = uma_zalloc(umtx_shm_reg_zone, M_WAITOK | M_ZERO);
4430 reg->ushm_refcnt = 1;
4431 bcopy(key, ®->ushm_key, sizeof(*key));
4432 reg->ushm_obj = shm_alloc(td->td_ucred, O_RDWR, false);
4433 reg->ushm_cred = crhold(cred);
4434 error = shm_dotruncate(reg->ushm_obj, PAGE_SIZE);
4435 if (error != 0) {
4436 umtx_shm_free_reg(reg);
4437 return (error);
4438 }
4439 mtx_lock(&umtx_shm_lock);
4440 reg1 = umtx_shm_find_reg_locked(key);
4441 if (reg1 != NULL) {
4442 mtx_unlock(&umtx_shm_lock);
4443 umtx_shm_free_reg(reg);
4444 *res = reg1;
4445 return (0);
4446 }
4447 reg->ushm_refcnt++;
4448 TAILQ_INSERT_TAIL(&umtx_shm_registry[key->hash], reg, ushm_reg_link);
4449 LIST_INSERT_HEAD(USHM_OBJ_UMTX(key->info.shared.object), reg,
4450 ushm_obj_link);
4451 reg->ushm_flags = USHMF_REG_LINKED | USHMF_OBJ_LINKED;
4452 mtx_unlock(&umtx_shm_lock);
4453 *res = reg;
4454 return (0);
4455 }
4456
4457 static int
4458 umtx_shm_alive(struct thread *td, void *addr)
4459 {
4460 vm_map_t map;
4461 vm_map_entry_t entry;
4462 vm_object_t object;
4463 vm_pindex_t pindex;
4464 vm_prot_t prot;
4465 int res, ret;
4466 boolean_t wired;
4467
4468 map = &td->td_proc->p_vmspace->vm_map;
4469 res = vm_map_lookup(&map, (uintptr_t)addr, VM_PROT_READ, &entry,
4470 &object, &pindex, &prot, &wired);
4471 if (res != KERN_SUCCESS)
4472 return (EFAULT);
4473 if (object == NULL)
4474 ret = EINVAL;
4475 else
4476 ret = (object->flags & OBJ_UMTXDEAD) != 0 ? ENOTTY : 0;
4477 vm_map_lookup_done(map, entry);
4478 return (ret);
4479 }
4480
4481 static void
4482 umtx_shm_init(void)
4483 {
4484 int i;
4485
4486 umtx_shm_reg_zone = uma_zcreate("umtx_shm", sizeof(struct umtx_shm_reg),
4487 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
4488 mtx_init(&umtx_shm_lock, "umtxshm", NULL, MTX_DEF);
4489 for (i = 0; i < nitems(umtx_shm_registry); i++)
4490 TAILQ_INIT(&umtx_shm_registry[i]);
4491 }
4492
4493 static int
4494 umtx_shm(struct thread *td, void *addr, u_int flags)
4495 {
4496 struct umtx_key key;
4497 struct umtx_shm_reg *reg;
4498 struct file *fp;
4499 int error, fd;
4500
4501 if (__bitcount(flags & (UMTX_SHM_CREAT | UMTX_SHM_LOOKUP |
4502 UMTX_SHM_DESTROY| UMTX_SHM_ALIVE)) != 1)
4503 return (EINVAL);
4504 if ((flags & UMTX_SHM_ALIVE) != 0)
4505 return (umtx_shm_alive(td, addr));
4506 error = umtx_key_get(addr, TYPE_SHM, PROCESS_SHARE, &key);
4507 if (error != 0)
4508 return (error);
4509 KASSERT(key.shared == 1, ("non-shared key"));
4510 if ((flags & UMTX_SHM_CREAT) != 0) {
4511 error = umtx_shm_create_reg(td, &key, ®);
4512 } else {
4513 reg = umtx_shm_find_reg(&key);
4514 if (reg == NULL)
4515 error = ESRCH;
4516 }
4517 umtx_key_release(&key);
4518 if (error != 0)
4519 return (error);
4520 KASSERT(reg != NULL, ("no reg"));
4521 if ((flags & UMTX_SHM_DESTROY) != 0) {
4522 umtx_shm_unref_reg(reg, true);
4523 } else {
4524 #if 0
4525 #ifdef MAC
4526 error = mac_posixshm_check_open(td->td_ucred,
4527 reg->ushm_obj, FFLAGS(O_RDWR));
4528 if (error == 0)
4529 #endif
4530 error = shm_access(reg->ushm_obj, td->td_ucred,
4531 FFLAGS(O_RDWR));
4532 if (error == 0)
4533 #endif
4534 error = falloc_caps(td, &fp, &fd, O_CLOEXEC, NULL);
4535 if (error == 0) {
4536 shm_hold(reg->ushm_obj);
4537 finit(fp, FFLAGS(O_RDWR), DTYPE_SHM, reg->ushm_obj,
4538 &shm_ops);
4539 td->td_retval[0] = fd;
4540 fdrop(fp, td);
4541 }
4542 }
4543 umtx_shm_unref_reg(reg, false);
4544 return (error);
4545 }
4546
4547 static int
4548 __umtx_op_shm(struct thread *td, struct _umtx_op_args *uap,
4549 const struct umtx_copyops *ops __unused)
4550 {
4551
4552 return (umtx_shm(td, uap->uaddr1, uap->val));
4553 }
4554
4555 static int
4556 __umtx_op_robust_lists(struct thread *td, struct _umtx_op_args *uap,
4557 const struct umtx_copyops *ops)
4558 {
4559 struct umtx_robust_lists_params rb;
4560 int error;
4561
4562 if (ops->compat32) {
4563 if ((td->td_pflags2 & TDP2_COMPAT32RB) == 0 &&
4564 (td->td_rb_list != 0 || td->td_rbp_list != 0 ||
4565 td->td_rb_inact != 0))
4566 return (EBUSY);
4567 } else if ((td->td_pflags2 & TDP2_COMPAT32RB) != 0) {
4568 return (EBUSY);
4569 }
4570
4571 bzero(&rb, sizeof(rb));
4572 error = ops->copyin_robust_lists(uap->uaddr1, uap->val, &rb);
4573 if (error != 0)
4574 return (error);
4575
4576 if (ops->compat32)
4577 td->td_pflags2 |= TDP2_COMPAT32RB;
4578
4579 td->td_rb_list = rb.robust_list_offset;
4580 td->td_rbp_list = rb.robust_priv_list_offset;
4581 td->td_rb_inact = rb.robust_inact_offset;
4582 return (0);
4583 }
4584
4585 #if defined(__i386__) || defined(__amd64__)
4586 /*
4587 * Provide the standard 32-bit definitions for x86, since native/compat32 use a
4588 * 32-bit time_t there. Other architectures just need the i386 definitions
4589 * along with their standard compat32.
4590 */
4591 struct timespecx32 {
4592 int64_t tv_sec;
4593 int32_t tv_nsec;
4594 };
4595
4596 struct umtx_timex32 {
4597 struct timespecx32 _timeout;
4598 uint32_t _flags;
4599 uint32_t _clockid;
4600 };
4601
4602 #ifndef __i386__
4603 #define timespeci386 timespec32
4604 #define umtx_timei386 umtx_time32
4605 #endif
4606 #else /* !__i386__ && !__amd64__ */
4607 /* 32-bit architectures can emulate i386, so define these almost everywhere. */
4608 struct timespeci386 {
4609 int32_t tv_sec;
4610 int32_t tv_nsec;
4611 };
4612
4613 struct umtx_timei386 {
4614 struct timespeci386 _timeout;
4615 uint32_t _flags;
4616 uint32_t _clockid;
4617 };
4618
4619 #if defined(__LP64__)
4620 #define timespecx32 timespec32
4621 #define umtx_timex32 umtx_time32
4622 #endif
4623 #endif
4624
4625 static int
4626 umtx_copyin_robust_lists32(const void *uaddr, size_t size,
4627 struct umtx_robust_lists_params *rbp)
4628 {
4629 struct umtx_robust_lists_params_compat32 rb32;
4630 int error;
4631
4632 if (size > sizeof(rb32))
4633 return (EINVAL);
4634 bzero(&rb32, sizeof(rb32));
4635 error = copyin(uaddr, &rb32, size);
4636 if (error != 0)
4637 return (error);
4638 CP(rb32, *rbp, robust_list_offset);
4639 CP(rb32, *rbp, robust_priv_list_offset);
4640 CP(rb32, *rbp, robust_inact_offset);
4641 return (0);
4642 }
4643
4644 #ifndef __i386__
4645 static inline int
4646 umtx_copyin_timeouti386(const void *uaddr, struct timespec *tsp)
4647 {
4648 struct timespeci386 ts32;
4649 int error;
4650
4651 error = copyin(uaddr, &ts32, sizeof(ts32));
4652 if (error == 0) {
4653 if (ts32.tv_sec < 0 ||
4654 ts32.tv_nsec >= 1000000000 ||
4655 ts32.tv_nsec < 0)
4656 error = EINVAL;
4657 else {
4658 CP(ts32, *tsp, tv_sec);
4659 CP(ts32, *tsp, tv_nsec);
4660 }
4661 }
4662 return (error);
4663 }
4664
4665 static inline int
4666 umtx_copyin_umtx_timei386(const void *uaddr, size_t size, struct _umtx_time *tp)
4667 {
4668 struct umtx_timei386 t32;
4669 int error;
4670
4671 t32._clockid = CLOCK_REALTIME;
4672 t32._flags = 0;
4673 if (size <= sizeof(t32._timeout))
4674 error = copyin(uaddr, &t32._timeout, sizeof(t32._timeout));
4675 else
4676 error = copyin(uaddr, &t32, sizeof(t32));
4677 if (error != 0)
4678 return (error);
4679 if (t32._timeout.tv_sec < 0 ||
4680 t32._timeout.tv_nsec >= 1000000000 || t32._timeout.tv_nsec < 0)
4681 return (EINVAL);
4682 TS_CP(t32, *tp, _timeout);
4683 CP(t32, *tp, _flags);
4684 CP(t32, *tp, _clockid);
4685 return (0);
4686 }
4687
4688 static int
4689 umtx_copyout_timeouti386(void *uaddr, size_t sz, struct timespec *tsp)
4690 {
4691 struct timespeci386 remain32 = {
4692 .tv_sec = tsp->tv_sec,
4693 .tv_nsec = tsp->tv_nsec,
4694 };
4695
4696 /*
4697 * Should be guaranteed by the caller, sz == uaddr1 - sizeof(_umtx_time)
4698 * and we're only called if sz >= sizeof(timespec) as supplied in the
4699 * copyops.
4700 */
4701 KASSERT(sz >= sizeof(remain32),
4702 ("umtx_copyops specifies incorrect sizes"));
4703
4704 return (copyout(&remain32, uaddr, sizeof(remain32)));
4705 }
4706 #endif /* !__i386__ */
4707
4708 #if defined(__i386__) || defined(__LP64__)
4709 static inline int
4710 umtx_copyin_timeoutx32(const void *uaddr, struct timespec *tsp)
4711 {
4712 struct timespecx32 ts32;
4713 int error;
4714
4715 error = copyin(uaddr, &ts32, sizeof(ts32));
4716 if (error == 0) {
4717 if (ts32.tv_sec < 0 ||
4718 ts32.tv_nsec >= 1000000000 ||
4719 ts32.tv_nsec < 0)
4720 error = EINVAL;
4721 else {
4722 CP(ts32, *tsp, tv_sec);
4723 CP(ts32, *tsp, tv_nsec);
4724 }
4725 }
4726 return (error);
4727 }
4728
4729 static inline int
4730 umtx_copyin_umtx_timex32(const void *uaddr, size_t size, struct _umtx_time *tp)
4731 {
4732 struct umtx_timex32 t32;
4733 int error;
4734
4735 t32._clockid = CLOCK_REALTIME;
4736 t32._flags = 0;
4737 if (size <= sizeof(t32._timeout))
4738 error = copyin(uaddr, &t32._timeout, sizeof(t32._timeout));
4739 else
4740 error = copyin(uaddr, &t32, sizeof(t32));
4741 if (error != 0)
4742 return (error);
4743 if (t32._timeout.tv_sec < 0 ||
4744 t32._timeout.tv_nsec >= 1000000000 || t32._timeout.tv_nsec < 0)
4745 return (EINVAL);
4746 TS_CP(t32, *tp, _timeout);
4747 CP(t32, *tp, _flags);
4748 CP(t32, *tp, _clockid);
4749 return (0);
4750 }
4751
4752 static int
4753 umtx_copyout_timeoutx32(void *uaddr, size_t sz, struct timespec *tsp)
4754 {
4755 struct timespecx32 remain32 = {
4756 .tv_sec = tsp->tv_sec,
4757 .tv_nsec = tsp->tv_nsec,
4758 };
4759
4760 /*
4761 * Should be guaranteed by the caller, sz == uaddr1 - sizeof(_umtx_time)
4762 * and we're only called if sz >= sizeof(timespec) as supplied in the
4763 * copyops.
4764 */
4765 KASSERT(sz >= sizeof(remain32),
4766 ("umtx_copyops specifies incorrect sizes"));
4767
4768 return (copyout(&remain32, uaddr, sizeof(remain32)));
4769 }
4770 #endif /* __i386__ || __LP64__ */
4771
4772 typedef int (*_umtx_op_func)(struct thread *td, struct _umtx_op_args *uap,
4773 const struct umtx_copyops *umtx_ops);
4774
4775 static const _umtx_op_func op_table[] = {
4776 #ifdef COMPAT_FREEBSD10
4777 [UMTX_OP_LOCK] = __umtx_op_lock_umtx,
4778 [UMTX_OP_UNLOCK] = __umtx_op_unlock_umtx,
4779 #else
4780 [UMTX_OP_LOCK] = __umtx_op_unimpl,
4781 [UMTX_OP_UNLOCK] = __umtx_op_unimpl,
4782 #endif
4783 [UMTX_OP_WAIT] = __umtx_op_wait,
4784 [UMTX_OP_WAKE] = __umtx_op_wake,
4785 [UMTX_OP_MUTEX_TRYLOCK] = __umtx_op_trylock_umutex,
4786 [UMTX_OP_MUTEX_LOCK] = __umtx_op_lock_umutex,
4787 [UMTX_OP_MUTEX_UNLOCK] = __umtx_op_unlock_umutex,
4788 [UMTX_OP_SET_CEILING] = __umtx_op_set_ceiling,
4789 [UMTX_OP_CV_WAIT] = __umtx_op_cv_wait,
4790 [UMTX_OP_CV_SIGNAL] = __umtx_op_cv_signal,
4791 [UMTX_OP_CV_BROADCAST] = __umtx_op_cv_broadcast,
4792 [UMTX_OP_WAIT_UINT] = __umtx_op_wait_uint,
4793 [UMTX_OP_RW_RDLOCK] = __umtx_op_rw_rdlock,
4794 [UMTX_OP_RW_WRLOCK] = __umtx_op_rw_wrlock,
4795 [UMTX_OP_RW_UNLOCK] = __umtx_op_rw_unlock,
4796 [UMTX_OP_WAIT_UINT_PRIVATE] = __umtx_op_wait_uint_private,
4797 [UMTX_OP_WAKE_PRIVATE] = __umtx_op_wake_private,
4798 [UMTX_OP_MUTEX_WAIT] = __umtx_op_wait_umutex,
4799 [UMTX_OP_MUTEX_WAKE] = __umtx_op_wake_umutex,
4800 #if defined(COMPAT_FREEBSD9) || defined(COMPAT_FREEBSD10)
4801 [UMTX_OP_SEM_WAIT] = __umtx_op_sem_wait,
4802 [UMTX_OP_SEM_WAKE] = __umtx_op_sem_wake,
4803 #else
4804 [UMTX_OP_SEM_WAIT] = __umtx_op_unimpl,
4805 [UMTX_OP_SEM_WAKE] = __umtx_op_unimpl,
4806 #endif
4807 [UMTX_OP_NWAKE_PRIVATE] = __umtx_op_nwake_private,
4808 [UMTX_OP_MUTEX_WAKE2] = __umtx_op_wake2_umutex,
4809 [UMTX_OP_SEM2_WAIT] = __umtx_op_sem2_wait,
4810 [UMTX_OP_SEM2_WAKE] = __umtx_op_sem2_wake,
4811 [UMTX_OP_SHM] = __umtx_op_shm,
4812 [UMTX_OP_ROBUST_LISTS] = __umtx_op_robust_lists,
4813 };
4814
4815 static const struct umtx_copyops umtx_native_ops = {
4816 .copyin_timeout = umtx_copyin_timeout,
4817 .copyin_umtx_time = umtx_copyin_umtx_time,
4818 .copyin_robust_lists = umtx_copyin_robust_lists,
4819 .copyout_timeout = umtx_copyout_timeout,
4820 .timespec_sz = sizeof(struct timespec),
4821 .umtx_time_sz = sizeof(struct _umtx_time),
4822 };
4823
4824 #ifndef __i386__
4825 static const struct umtx_copyops umtx_native_opsi386 = {
4826 .copyin_timeout = umtx_copyin_timeouti386,
4827 .copyin_umtx_time = umtx_copyin_umtx_timei386,
4828 .copyin_robust_lists = umtx_copyin_robust_lists32,
4829 .copyout_timeout = umtx_copyout_timeouti386,
4830 .timespec_sz = sizeof(struct timespeci386),
4831 .umtx_time_sz = sizeof(struct umtx_timei386),
4832 .compat32 = true,
4833 };
4834 #endif
4835
4836 #if defined(__i386__) || defined(__LP64__)
4837 /* i386 can emulate other 32-bit archs, too! */
4838 static const struct umtx_copyops umtx_native_opsx32 = {
4839 .copyin_timeout = umtx_copyin_timeoutx32,
4840 .copyin_umtx_time = umtx_copyin_umtx_timex32,
4841 .copyin_robust_lists = umtx_copyin_robust_lists32,
4842 .copyout_timeout = umtx_copyout_timeoutx32,
4843 .timespec_sz = sizeof(struct timespecx32),
4844 .umtx_time_sz = sizeof(struct umtx_timex32),
4845 .compat32 = true,
4846 };
4847
4848 #ifdef COMPAT_FREEBSD32
4849 #ifdef __amd64__
4850 #define umtx_native_ops32 umtx_native_opsi386
4851 #else
4852 #define umtx_native_ops32 umtx_native_opsx32
4853 #endif
4854 #endif /* COMPAT_FREEBSD32 */
4855 #endif /* __i386__ || __LP64__ */
4856
4857 #define UMTX_OP__FLAGS (UMTX_OP__32BIT | UMTX_OP__I386)
4858
4859 static int
4860 kern__umtx_op(struct thread *td, void *obj, int op, unsigned long val,
4861 void *uaddr1, void *uaddr2, const struct umtx_copyops *ops)
4862 {
4863 struct _umtx_op_args uap = {
4864 .obj = obj,
4865 .op = op & ~UMTX_OP__FLAGS,
4866 .val = val,
4867 .uaddr1 = uaddr1,
4868 .uaddr2 = uaddr2
4869 };
4870
4871 if ((uap.op >= nitems(op_table)))
4872 return (EINVAL);
4873 return ((*op_table[uap.op])(td, &uap, ops));
4874 }
4875
4876 int
4877 sys__umtx_op(struct thread *td, struct _umtx_op_args *uap)
4878 {
4879 static const struct umtx_copyops *umtx_ops;
4880
4881 umtx_ops = &umtx_native_ops;
4882 #ifdef __LP64__
4883 if ((uap->op & (UMTX_OP__32BIT | UMTX_OP__I386)) != 0) {
4884 if ((uap->op & UMTX_OP__I386) != 0)
4885 umtx_ops = &umtx_native_opsi386;
4886 else
4887 umtx_ops = &umtx_native_opsx32;
4888 }
4889 #elif !defined(__i386__)
4890 /* We consider UMTX_OP__32BIT a nop on !i386 ILP32. */
4891 if ((uap->op & UMTX_OP__I386) != 0)
4892 umtx_ops = &umtx_native_opsi386;
4893 #else
4894 /* Likewise, UMTX_OP__I386 is a nop on i386. */
4895 if ((uap->op & UMTX_OP__32BIT) != 0)
4896 umtx_ops = &umtx_native_opsx32;
4897 #endif
4898 return (kern__umtx_op(td, uap->obj, uap->op, uap->val, uap->uaddr1,
4899 uap->uaddr2, umtx_ops));
4900 }
4901
4902 #ifdef COMPAT_FREEBSD32
4903 #ifdef COMPAT_FREEBSD10
4904 int
4905 freebsd10_freebsd32_umtx_lock(struct thread *td,
4906 struct freebsd10_freebsd32_umtx_lock_args *uap)
4907 {
4908 return (do_lock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid, NULL));
4909 }
4910
4911 int
4912 freebsd10_freebsd32_umtx_unlock(struct thread *td,
4913 struct freebsd10_freebsd32_umtx_unlock_args *uap)
4914 {
4915 return (do_unlock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid));
4916 }
4917 #endif /* COMPAT_FREEBSD10 */
4918
4919 int
4920 freebsd32__umtx_op(struct thread *td, struct freebsd32__umtx_op_args *uap)
4921 {
4922
4923 return (kern__umtx_op(td, uap->obj, uap->op, uap->val, uap->uaddr,
4924 uap->uaddr2, &umtx_native_ops32));
4925 }
4926 #endif /* COMPAT_FREEBSD32 */
4927
4928 void
4929 umtx_thread_init(struct thread *td)
4930 {
4931
4932 td->td_umtxq = umtxq_alloc();
4933 td->td_umtxq->uq_thread = td;
4934 }
4935
4936 void
4937 umtx_thread_fini(struct thread *td)
4938 {
4939
4940 umtxq_free(td->td_umtxq);
4941 }
4942
4943 /*
4944 * It will be called when new thread is created, e.g fork().
4945 */
4946 void
4947 umtx_thread_alloc(struct thread *td)
4948 {
4949 struct umtx_q *uq;
4950
4951 uq = td->td_umtxq;
4952 uq->uq_inherited_pri = PRI_MAX;
4953
4954 KASSERT(uq->uq_flags == 0, ("uq_flags != 0"));
4955 KASSERT(uq->uq_thread == td, ("uq_thread != td"));
4956 KASSERT(uq->uq_pi_blocked == NULL, ("uq_pi_blocked != NULL"));
4957 KASSERT(TAILQ_EMPTY(&uq->uq_pi_contested), ("uq_pi_contested is not empty"));
4958 }
4959
4960 /*
4961 * exec() hook.
4962 *
4963 * Clear robust lists for all process' threads, not delaying the
4964 * cleanup to thread exit, since the relevant address space is
4965 * destroyed right now.
4966 */
4967 void
4968 umtx_exec(struct proc *p)
4969 {
4970 struct thread *td;
4971
4972 KASSERT(p == curproc, ("need curproc"));
4973 KASSERT((p->p_flag & P_HADTHREADS) == 0 ||
4974 (p->p_flag & P_STOPPED_SINGLE) != 0,
4975 ("curproc must be single-threaded"));
4976 /*
4977 * There is no need to lock the list as only this thread can be
4978 * running.
4979 */
4980 FOREACH_THREAD_IN_PROC(p, td) {
4981 KASSERT(td == curthread ||
4982 ((td->td_flags & TDF_BOUNDARY) != 0 && TD_IS_SUSPENDED(td)),
4983 ("running thread %p %p", p, td));
4984 umtx_thread_cleanup(td);
4985 td->td_rb_list = td->td_rbp_list = td->td_rb_inact = 0;
4986 }
4987 }
4988
4989 /*
4990 * thread exit hook.
4991 */
4992 void
4993 umtx_thread_exit(struct thread *td)
4994 {
4995
4996 umtx_thread_cleanup(td);
4997 }
4998
4999 static int
5000 umtx_read_uptr(struct thread *td, uintptr_t ptr, uintptr_t *res, bool compat32)
5001 {
5002 u_long res1;
5003 uint32_t res32;
5004 int error;
5005
5006 if (compat32) {
5007 error = fueword32((void *)ptr, &res32);
5008 if (error == 0)
5009 res1 = res32;
5010 } else {
5011 error = fueword((void *)ptr, &res1);
5012 }
5013 if (error == 0)
5014 *res = res1;
5015 else
5016 error = EFAULT;
5017 return (error);
5018 }
5019
5020 static void
5021 umtx_read_rb_list(struct thread *td, struct umutex *m, uintptr_t *rb_list,
5022 bool compat32)
5023 {
5024 struct umutex32 m32;
5025
5026 if (compat32) {
5027 memcpy(&m32, m, sizeof(m32));
5028 *rb_list = m32.m_rb_lnk;
5029 } else {
5030 *rb_list = m->m_rb_lnk;
5031 }
5032 }
5033
5034 static int
5035 umtx_handle_rb(struct thread *td, uintptr_t rbp, uintptr_t *rb_list, bool inact,
5036 bool compat32)
5037 {
5038 struct umutex m;
5039 int error;
5040
5041 KASSERT(td->td_proc == curproc, ("need current vmspace"));
5042 error = copyin((void *)rbp, &m, sizeof(m));
5043 if (error != 0)
5044 return (error);
5045 if (rb_list != NULL)
5046 umtx_read_rb_list(td, &m, rb_list, compat32);
5047 if ((m.m_flags & UMUTEX_ROBUST) == 0)
5048 return (EINVAL);
5049 if ((m.m_owner & ~UMUTEX_CONTESTED) != td->td_tid)
5050 /* inact is cleared after unlock, allow the inconsistency */
5051 return (inact ? 0 : EINVAL);
5052 return (do_unlock_umutex(td, (struct umutex *)rbp, true));
5053 }
5054
5055 static void
5056 umtx_cleanup_rb_list(struct thread *td, uintptr_t rb_list, uintptr_t *rb_inact,
5057 const char *name, bool compat32)
5058 {
5059 int error, i;
5060 uintptr_t rbp;
5061 bool inact;
5062
5063 if (rb_list == 0)
5064 return;
5065 error = umtx_read_uptr(td, rb_list, &rbp, compat32);
5066 for (i = 0; error == 0 && rbp != 0 && i < umtx_max_rb; i++) {
5067 if (rbp == *rb_inact) {
5068 inact = true;
5069 *rb_inact = 0;
5070 } else
5071 inact = false;
5072 error = umtx_handle_rb(td, rbp, &rbp, inact, compat32);
5073 }
5074 if (i == umtx_max_rb && umtx_verbose_rb) {
5075 uprintf("comm %s pid %d: reached umtx %smax rb %d\n",
5076 td->td_proc->p_comm, td->td_proc->p_pid, name, umtx_max_rb);
5077 }
5078 if (error != 0 && umtx_verbose_rb) {
5079 uprintf("comm %s pid %d: handling %srb error %d\n",
5080 td->td_proc->p_comm, td->td_proc->p_pid, name, error);
5081 }
5082 }
5083
5084 /*
5085 * Clean up umtx data.
5086 */
5087 static void
5088 umtx_thread_cleanup(struct thread *td)
5089 {
5090 struct umtx_q *uq;
5091 struct umtx_pi *pi;
5092 uintptr_t rb_inact;
5093 bool compat32;
5094
5095 /*
5096 * Disown pi mutexes.
5097 */
5098 uq = td->td_umtxq;
5099 if (uq != NULL) {
5100 if (uq->uq_inherited_pri != PRI_MAX ||
5101 !TAILQ_EMPTY(&uq->uq_pi_contested)) {
5102 mtx_lock(&umtx_lock);
5103 uq->uq_inherited_pri = PRI_MAX;
5104 while ((pi = TAILQ_FIRST(&uq->uq_pi_contested)) != NULL) {
5105 pi->pi_owner = NULL;
5106 TAILQ_REMOVE(&uq->uq_pi_contested, pi, pi_link);
5107 }
5108 mtx_unlock(&umtx_lock);
5109 }
5110 sched_lend_user_prio_cond(td, PRI_MAX);
5111 }
5112
5113 compat32 = (td->td_pflags2 & TDP2_COMPAT32RB) != 0;
5114 td->td_pflags2 &= ~TDP2_COMPAT32RB;
5115
5116 if (td->td_rb_inact == 0 && td->td_rb_list == 0 && td->td_rbp_list == 0)
5117 return;
5118
5119 /*
5120 * Handle terminated robust mutexes. Must be done after
5121 * robust pi disown, otherwise unlock could see unowned
5122 * entries.
5123 */
5124 rb_inact = td->td_rb_inact;
5125 if (rb_inact != 0)
5126 (void)umtx_read_uptr(td, rb_inact, &rb_inact, compat32);
5127 umtx_cleanup_rb_list(td, td->td_rb_list, &rb_inact, "", compat32);
5128 umtx_cleanup_rb_list(td, td->td_rbp_list, &rb_inact, "priv ", compat32);
5129 if (rb_inact != 0)
5130 (void)umtx_handle_rb(td, rb_inact, NULL, true, compat32);
5131 }
Cache object: f4583d6772ebd96651d52f777e28f689
|